How to Find Comparable Sales for a House You Want to Buy

## What a comparable sale actually is

USPAP — the Uniform Standards of Professional Appraisal Practice
that every licensed US appraiser follows — defines the
sales-comparison approach as the analysis of "recently sold
properties similar to the subject in characteristics relevant to
value." That word *similar* is doing a lot of work. In practice
USPAP and the federally-recognised appraisal standards (Fannie
Mae's UAD, Freddie Mac's UCDP) require comps that share the
target's property type, are within a defensible geographic
radius, sold within the past 12 months at the outside, and were
arm's-length transactions — meaning a willing buyer paid a willing
seller without coercion or insider relationships.

The buyer-side definition is more direct. A comp is a sold house
that you can point at and say: "if I weren't bidding on this
specific property, *that* one is what my money would have bought
me last quarter." The closer the match — same street, same
floorplan, same vintage, sold last month — the more weight that
sale carries in your offer math.

The trap most first-time buyers fall into is using listings
that are still **active** or **pending** as comps. Those aren't
comps; those are asking prices. A house listed at $499,000 tells
you nothing about what it will sell for. A house *that closed*
at $478,500 tells you a contract was actually written and a
lender actually funded the loan against that number. Only closed
sales are evidence; everything else is opinion.

## Where to find comparable sales

There is no single national database of US home sales the public
can query. There are roughly 600 regional MLSs, fifty state
recording systems, and a handful of consumer aggregators that
licence and republish that data. Here is the practical pecking
order for a buyer working an address in 2026:

| Source | Coverage | Speed | What it shows | Cost | Catch |
|---|---|---|---|---|---|
| **Local MLS** (via agent or flat-fee portal) | ~95% of arm's-length sales | Same-day | Full listing, photos, DOM, price history, concessions | Agent commission or $50–$200 flat fee | Restricted to licensed members; non-members get filtered IDX feeds |
| **Redfin "Recently Sold"** | ~85–90% of MLS | 1–7 days lag | Closed price, beds/baths/sqft, photos, basic timeline | Free | "Off-market estimate" is a model, not the recorded sale price |
| **Zillow "Sold Homes" filter** | ~90% (regional gaps) | 1–10 days lag | Closed price, photo set, Zestimate history, school zone | Free | Some MLSs (e.g. parts of NJ, NC) suppress Zillow feeds |
| **County recorder / assessor** | 100% of deed transfers | 1–6 weeks lag | Buyer/seller names, sale price, deed type, parcel ID | Free or $1–$5 per record | Misses concessions, no photos, no listing context |
| **FHFA House Price Index** | National repeat-sales | Quarterly | Trend / appreciation rate by metro | Free | Trend only, never individual sales |
| **Twellie report** | MLS + recorder fused | Same-day | Top 8 ranked comps with adjustments + photo grades | $50 per address | One address per pull |

The four-step buyer playbook: start with **Redfin** to get a
visual feel for the neighborhood and an unfiltered list of
recently-sold homes you can click through, cross-check anything
that looks unusual against the **county recorder** (free, and
the only source that catches non-MLS sales like estate
transfers and FSBOs), use **FHFA's repeat-sales index** to set
your time-of-sale adjustment for the local trend, and then —
when you're serious about an address — pull a **Twellie report**
or have your buyer's agent run a CMA off the live MLS.

The MLS is still the highest-quality feed because it captures
two fields the other sources miss: **days-on-market** and
**seller concessions**. A house that closed at $478,500 with
$8,000 in seller-paid closing costs effectively cleared $470,500
— a four-figure adjustment that quietly moves your offer math.
You will not see that field on Redfin or Zillow. You will see it
on the MLS, and Twellie pulls it through.

For sale-price verification on any deal that feels weird, the
county recorder is the source of truth. The recorded deed is the
legal document; if Zillow says $510,000 and the deed says
$485,000, the deed is right. This is also where you catch
non-arm's-length sales — quitclaim deeds, transfers between
family members, foreclosure-trustee deeds — that should be
excluded from your comp set.

## How to filter the candidate set

Once you have a candidate list — typically 30 to 200 sold homes
within a generous starting box — you tighten it with three
filters: **distance**, **time**, and **property similarity**.

### Distance (radius)

Real estate value is hyperlocal. The right radius depends on
density, not on your city's name.

* **Dense urban (Manhattan, SF, Boston Back Bay)** —
  0.10 to 0.25 miles. A six-block walk can cross a $200/ft²
  micromarket boundary. Comps from the next neighborhood over
  belong in a different report.
* **Standard urban / inner suburban** — 0.25 to 0.5 miles.
  This covers the typical American walkable neighborhood and
  matches what a Fannie Mae-compliant appraiser will use.
* **Suburban single-family** — 0.5 to 1.0 miles. School-zone
  boundaries matter more than literal distance; a comp 0.3
  miles away in a different elementary catchment is *less*
  comparable than one 0.8 miles away in the same catchment.
* **Rural / large-lot** — 1.0 to 5.0 miles, sometimes more.
  In rural Vermont or West Texas you may be expanding to the
  same county or even the same school district. The radius
  is whatever it takes to find five sold homes of the same
  general type.

A practical rule: if your radius is more than 1.0 mile in a
suburban setting or more than 0.5 in an urban one, you are no
longer measuring the local market — you are interpolating.
Widen the *time* window before you widen the *distance* window.

### Time window

How recently does a sale need to have closed to count as a comp?

* **Hot, fast-moving market (>5% annual appreciation):**
  90 days max. Markets at this pace can move 1% per month;
  a six-month-old comp is already a 6% adjustment, which is
  too much to ride on.
* **Normal market (0–5% annual):** 180 days. The Fannie Mae
  default for residential appraisals is six months, with up
  to twelve allowed if explained.
* **Declining or thin market:** Up to 12 months, and use
  pendings as a sanity check on direction. In a falling
  market, older comps overstate value.
* **Rural / very low volume:** Whatever the market gives you.
  In a market with one sale per quarter you have no choice
  but to use a 9-month-old comp; just expect the confidence
  band on your AVM to be wide and price your offer humbly.

The single biggest mistake amateur buyers make is treating a
comp from "earlier this year" as current evidence. Eight months
in a 0.6%-per-month market is a 4.8% adjustment — close to
$25,000 on a $500,000 home. Read more on the date-of-sale
adjustment in our companion guide,
[/guides/comp-adjustment-factors-explained](/guides/comp-adjustment-factors-explained).

### Property similarity

The third filter is the one most buyers tighten too aggressively.
Counterintuitively, **looser is sometimes better** — a comp
that's slightly different but extremely recent and extremely
close is usually more useful than a perfect twin from 18 months
and three miles away.

Reasonable similarity bounds for a typical single-family search:

* **Property type:** identical. A condo is never a comp for a
  single-family detached. A townhome is never a comp for a
  duplex. This is non-negotiable.
* **Square footage:** target ±20%. A 2,000 ft² house can be
  comp'd to homes from 1,600 to 2,400. Outside that band the
  hedonic regression starts breaking down.
* **Bedrooms:** ±1. A 4-bed can take 3- and 5-bed comps; a
  3-bed taking 1-bed comps is a stretch.
* **Bathrooms:** ±1 full bath equivalent (with halves counted
  at 0.5).
* **Year built:** ±15 years for typical post-war stock, ±25
  years for older urban. A 1915 craftsman and a 2018 new build
  are not comps even on the same street.
* **Lot size:** ±50% in suburban; type-equivalent in urban
  (no half-acre comps for a townhome on a 1,500 ft² lot).

## What makes a comp "good": the six quality criteria

Once filtered, every remaining candidate should be scored
against six criteria that human appraisers and modern AVMs both
weigh.

1. **Arm's-length sale.** A willing buyer paid a willing seller,
   neither under duress, neither related. Foreclosure auctions,
   estate transfers, family quitclaims, and corporate-relocation
   buyouts are not arm's-length and don't belong in a comp set.
   Look for the deed type on the county recorder — *warranty
   deed* is good; *trustee's deed*, *quitclaim*, or *executor's
   deed* require investigation.

2. **Standard financing or cash, not unusual concessions.** A
   sale with $25,000 in seller credits, a 3% rate buydown, or a
   below-market lease-back to the seller is not a clean signal.
   Either net those concessions out of the price or exclude the
   comp.

3. **No major condition mismatch the photos can't resolve.**
   A renovated comp on the same block as an unrenovated target
   doesn't tell you what the target is worth — it tells you
   what the *renovated* version is worth. Use it, but apply the
   condition adjustment honestly. Twellie's photo-condition
   scoring grades each visible room A–F and applies the
   adjustment automatically.

4. **Adequate photos and listing detail.** A comp with three
   photos and a 40-word description is half-evidence. You can
   still use it, but its weight in your average should be
   reduced. A comp with 35 photos, a video walkthrough, and a
   detailed remarks section is high-confidence.

5. **Reasonable days-on-market.** A comp that closed in 4 days
   tells you the market is hot at that price point. A comp that
   sat for 110 days and closed at 92% of asking tells you the
   opposite. Use DOM as a tie-breaker — recent fast sales weight
   higher than recent slow sales for the *direction* of the
   market.

6. **Verified sale price.** Cross-check the MLS-reported close
   price against the recorded deed. They should match (within
   any explicit concessions). When they don't, trust the deed.
   Discrepancies usually mean undisclosed seller credits.

The standard appraisal-industry shorthand for this is the
**comparability score** — a 0-to-100 (or 1-to-10) summary of
how close the comp is to the target across all six criteria.
Twellie computes a similarity score across 14 attributes —
including the six above plus walkability, school catchment,
flood-zone match, HOA status, and view orientation — and ranks
the candidate pool. The top eight ranked comps land in the
report, with the score visible next to each.

## How many comps do you actually need?

* **Three** is the absolute floor. Below three, you have
  anecdotes, not a market signal. Federal appraisal guidelines
  (UAD form 1004) require at least three comps for a
  conventional residential appraisal — that's the legal
  minimum and it exists for a statistical reason.
* **Five to seven** is the working sweet spot for a buyer's
  CMA. With five comps you start washing out the noise from
  any single weird transaction. With seven you have enough
  data to spot outliers and exclude them.
* **Eight to ten** is what a thorough AVM uses internally
  before averaging. Twellie's report shows the top eight after
  the engine considers 50–200 candidates.
* **More than ten** dilutes signal. The 11th-best comp is, by
  definition, less similar than the top three — adding it
  pulls the central estimate toward irrelevant transactions.

The math: as you average more comps, the standard error of the
mean falls, but only as 1/√n. Going from 3 to 7 comps roughly
halves the noise; going from 7 to 30 only halves it again. The
diminishing return shows up around comp #7, which is why most
appraisers and AVMs stop there.

## Reading the comp set: spotting outliers and signal

Once you have 5–7 comps lined up, plot them. The shape of the
distribution tells you almost as much as the average.

**Tight cluster.** Five comps sold within a $20,000 range on a
$450,000 home. The market signal is loud and clean — your offer
should anchor near the middle of the cluster. Confidence band
on the AVM will be narrow.

**Wide cluster.** Five comps spanning $385,000 to $510,000 on
the same target. Either the comp set is genuinely
heterogeneous (different conditions, different floorplans,
different micro-locations) or the market itself is in
transition. Tighten the similarity filter and re-pull.

**One outlier.** Four comps at $440,000–$465,000 and one at
$525,000. Investigate the outlier. Most often it's a flipped
home with a renovation the others didn't have, or a buyer who
overpaid in a multi-offer situation. Either way, downweight or
exclude.

**Bimodal distribution.** Two comps at the low end, two at the
high end, no middle. This is a common signature of a
neighborhood with mixed vintages — say, original 1960s ranchers
selling next to fully-rehabbed flips. The condition adjustment
is doing more work than the location adjustment in this case,
and a photo-aware AVM (or a careful manual review) will
materially out-perform a public Zestimate that ignores condition.
This is the gap that makes paid AVMs like Twellie or HouseCanary
worth their fee on mixed-stock streets — see
[/guides/avm-vs-appraisal-vs-zestimate](/guides/avm-vs-appraisal-vs-zestimate)
for the accuracy benchmarks behind that claim.

## Common mistakes when picking comps

**Mixing HOA and non-HOA properties.** A $300/month HOA
capitalizes to roughly $50,000 of price differential at
prevailing cap rates. Mixing condo or PUD comps with
non-HOA single-family at the same nominal sqft is one of the
most expensive errors in amateur comp work.

**Using golf-course frontage comps for an interior lot.** View,
backing-to-amenity, and corner-lot premiums can be 5–20% of
value. A comp on the 9th hole and a comp on the cul-de-sac
across the street are not interchangeable inside the same
subdivision.

**Including a flip without adjusting.** A house bought distressed
six months ago for $310,000, renovated, and resold today for
$485,000 is a flipped sale. The $485,000 comp reflects post-flip
finish quality. If the target hasn't been touched since 1998,
that comp overstates value by the renovation budget — typically
$40,000–$120,000.

**Using the same comp twice through different sources.** A sale
that appears on the MLS, on Redfin, on Zillow, and in the
county records is *one* sale, not four data points. De-duplicate
on parcel ID before averaging.

**Treating active listings as comps.** Worth repeating. Asking
prices are not market evidence. Only closed prices count.

**Ignoring concessions.** A $500,000 closed price with $15,000
in seller-paid closing costs is a $485,000 net sale. The MLS
field is small and most amateurs miss it.

## What to do next

Pull a Twellie report on the address you're seriously
considering — the report runs 50–200 comp candidates through
the similarity engine, surfaces the top eight, shows the
adjustments line-by-line, and translates the comp evidence into
a recommended offer range with a confidence band. The
methodology behind the comp engine is documented at
[/methodology](/methodology), and a complete sample report —
including the comp section — is at [/mockup/report](/mockup/report).

For the second half of comp work — what adjustments to apply
*after* you have the right comps — read our companion guide
[/guides/comp-adjustment-factors-explained](/guides/comp-adjustment-factors-explained).
For everything else in a valuation report (confidence bands,
photo grades, cost-of-ownership, risk profile), start with
[/guides/how-to-read-a-home-valuation-report](/guides/how-to-read-a-home-valuation-report).

Comp finding is half art, half discipline. The art is knowing
when a 0.4-mile comp from last month beats a next-door comp
from last year; the discipline is filtering ruthlessly and
verifying every closed price against the recorded deed. Get
both halves right and you'll write offers anchored to evidence
the seller can't argue with — which is the only kind of offer
that wins without overpaying.