Finding good comps is the most important skill in real estate underwriting. Everything downstream — ARV, MAO, offer price, buyer package — depends on starting with the right sales data. A great formula with bad comps gives you a precise wrong answer.
This is your checklist, plus where to actually get the data if you don't have MLS access.
The quality checklist: what makes a comp good
Good comps pass three filters, applied in this order:
Location. The comp should be close to your subject property and on the same side of any major boundary: highway, railroad, school district line, flood plain boundary. A half-mile radius is the starting default for most suburban and urban markets. Rural markets may need a larger radius, but the boundary principle still applies.
If you don't know the local market, look at where prices cluster on a map of recent sales. The invisible lines usually become visible when you look at enough data — you'll see a price drop or jump that corresponds to some physical or administrative boundary you didn't know about.
Recency. Use sales from the last 90 days whenever possible. In thin markets, extend to 6 months. Beyond 6 months, you're looking at a market that may have been meaningfully different — different interest rate environment, different inventory levels, different seasonal dynamics.
If you're forced to use older comps, explicitly note the time gap and adjust for any known market direction. "The market has appreciated about 3% over the past year, so I adjusted a 10-month-old sale up by 2.5%" is a defensible note. An unadjusted 18-month-old comp is not.
Property fit. Same general beds and baths (within one of each). Square footage within about 15-20% of your subject. Similar age bracket — decade-built is usually fine, but a 1960 house doesn't comp the same as a 2002 house even if they're the same size. Similar condition and renovation level: renovated comps for ARV estimation, as-is comps for current-condition value.
The disqualification checklist: what to remove
Once you have a pool of candidates, screen for these:
Non-arm's-length sales. Transactions between family members, related LLCs, or estates below market. The tell: sold price far below neighboring comparables, or very short DOM.
Foreclosure and REO sales. These often trade at a significant discount to what a willing seller would accept. They represent a distressed market, not a retail market. Unless your subject is also distressed and your buyer is an investor, these don't belong in your ARV calculation.
Outliers. A sale significantly above or below the cluster of the others. Investigate before using it — it might represent a premium for something your subject doesn't have (a pool, a view, a significantly larger lot) or a distressed condition.
Sales across a boundary. Even if the sale is close in distance, check the school district, flood zone, and major road separations. Being 0.3 miles away but in a different school district may mean a meaningfully different buyer pool and price point.
Where to get comp data
If you have MLS access: Use the MLS directly. Filter for sold status, set your date range, draw your geography, and match property type. This is the gold standard.
If you don't have MLS access:
Redfin.com lets you view recently sold properties in any area, and you can set a date filter. Not every state's MLS feeds into Redfin equally, but it's the most complete free option.
Zillow has sold data under the "sold" filter, though it's less reliable in some markets and the data quality varies by region.
County assessor/recorder websites show recorded deed transfers with prices in most states. Slower to navigate, but official. Useful as a cross-check.
Realtor.com has sold data and typically pulls from local MLS feeds.
Purpose-built platforms like Underwritey pull sold data automatically when you run a property, filtered and ranked by the criteria above. This is faster than manual research across multiple sites.
A note on Zillow's Zestimate: It's not a comp. The Zestimate is an automated valuation model that aggregates data from many sources and applies a statistical model. It's useful for quick ballpark orientation, not for underwriting decisions that involve real money.
How many comps do you need?
Three strong comps are better than eight mediocre ones. Your minimum is three; five is a good target. Beyond six, you're usually including comps that are further away, older, or less similar — they add statistical noise rather than accuracy.
When the market is truly thin and you can only find one or two qualifying sales, be explicit about it: document the limitation and note that the ARV has higher uncertainty than usual. Don't stretch to include bad comps just to hit a magic number.
Applying the checklist consistently
The discipline of applying the same checklist to every deal protects you from motivated reasoning — the tendency to include comps that support the price you want and exclude ones that don't. Run the filters first, then see what ARV the qualifying sales produce.
If you don't like the ARV the data produces, that's important information. The market is telling you something about what the property is actually worth. Adjust your repair estimate, adjust your strategy, or walk away — but don't adjust the comps to fit your desired number.
Underwritey runs this checklist automatically on every address, ranking comps by location and recency first, then fit. You review the results and remove anything that doesn't belong. Start your first underwrite for free →