How to run comps like an appraiser, with or without MLS access

Comps decide your offer, your list price and your assignment fee, and most people run them backwards. They open a map, sort by nearby sales, and grab the three biggest numbers. This is the method a licensed appraiser uses instead, why two of the rules you have read online are no longer true, and how to get to a defensible number with or without an MLS login.

What a comp actually is

A comp is one recently sold property close enough to yours in location, size, age and condition that its price tells you what yours is worth. Sold. Not listed, not pending, not an online estimate. A listing is an asking price, which is a hope. A closed sale is a number two parties signed.

The job a comp does is narrow. It sets the price for one decision. A flipper backs a maximum offer out of it, a wholesaler prices an assignment on it, a listing agent picks a number the seller will sign and a buyer will pay. Get the comps wrong and every number downstream is wrong with them, which is the whole reason to run them by hand at least once before you trust software to do it. Where the comping ends, the offer math begins, and our guide to calculating ARV takes it from there.

One caution first. This is general method, not an appraisal and not financial advice. For a loan, a tax appeal or a court case, consult a licensed appraiser, and remember your own street can behave differently than any national rule, so treat what follows as a frame you check against your local sold data rather than gospel.

Start with the subject, not the sold list

You cannot judge whether a sale is comparable until you have written down what you are comparing it to. So pull the subject property first, before you look at a single price.

Write down the gross living area in square feet, the bedroom and bathroom count, the lot size, the year built and the garage spaces. Then the variable that wrecks amateur comps: condition. A 1,600-square-foot ranch gutted and re-plumbed in 2024 is not the same asset as the identical shell next door with a 1988 kitchen and a roof at the end of its life, even though the county tax card lists them as twins. Read the condition of your subject now and put it in one honest line, while no sold price is around to bend your judgment.

How appraisers pick comparables

The appraisal industry solved this problem decades ago, and the rulebook is public. Fannie Mae's Selling Guide is the document every conventional-loan appraiser in the country works to. It requires a minimum of three closed comparable sales in the sales comparison approach, and it says sales that closed within the last 12 months should be used. That guidance took effect on June 4, 2025, and it is still current.

Now read what it does not say. There is no one-mile rule. Fannie tells the appraiser to stay inside the subject's market area and to state the exact distance and direction, so a real report reads "1.75 miles NW," never "close by." In a dense grid of near-identical homes your best comp might sit four doors down. Across a ravine, a highway or a school-district boundary it might not exist at half a mile. Distance is only a stand-in for similarity. The second you have a better stand-in, use it.

Good appraisers also record every candidate on Fannie's sales comparison grid and bracket the subject. They pick some sales clearly better than the subject and some clearly worse, so the value lands inside a range pinned on both ends instead of dangling off one side. Three comps that are all nicer than your house tell you a ceiling and nothing else. The filter below is what every candidate sale has to pass.

The adjustment grid, and the rule that no longer exists

No two houses match exactly, so you adjust. If a comp has a garage and your subject does not, you subtract the market value of that garage from the comp's price to line them up. An extra bathroom, a finished basement, another 300 square feet: each gets a dollar figure, and the adjustment always lands on the comp, never on the subject.

Here is where most guides are flatly wrong. They tell you an appraiser cannot make net adjustments over 15 percent or gross adjustments over 25 percent, and that you should throw out any comp that breaks those caps. That rule was retired. Fannie's current guidance states in plain words that it "does not have specific limitations or guidelines associated with net or gross adjustments," and that the number or size of adjustments cannot be the only reason to reject a comp. What counts is whether each adjustment reflects how buyers in that market actually price the difference. A page still quoting the 15-and-25 rule as law in 2026 is quoting a guideline Fannie deleted.

You will not own an appraiser's paired-sales database, and you do not need one. Use conservative, market-based numbers you can defend to a skeptic: a bathroom is worth what two otherwise-matched sales say it is worth in your zip, not a figure you found in a national article. When you are unsure, adjust less. A smaller, honest adjustment beats a big guess dressed up as precision.

Running comps without MLS access

No license and no MLS still gets you to a defensible number, because closed sales are public across most of the country. Two records carry the weight. The county recorder, or register of deeds, holds the deed, which shows the sale date and, in most states, the price. The county assessor holds the property card with square footage, room counts, year built and lot size. Between them you can rebuild the raw material an appraiser buys from a data vendor, for free, one address at a time.

The consumer portals sit on top of those same records. On Zillow and Redfin, filter to Sold, set a tight date window, and match beds, baths and square footage to your subject; Redfin often posts sold data quickly because it is a brokerage wired into MLS feeds. Paid tools stack the recorder, assessor and MLS data into one screen and save the clicking, which is the case for something like PropStream, and we lay out the tradeoffs in our PropStream comparison. Whatever you use, you are still checking the same property data by hand at the end.

One trap bites in exactly the states where the public-record method breaks. In non-disclosure states, roughly a dozen of them including Texas, Missouri, Kansas, Montana and New Mexico, the sale price is not part of the public record at all. You can see that a house on that street sold in March and never see the figure. There, the free recorder route stalls, and you fall back to MLS-fed estimates, an agent who will pull sold comps for you, or paid data. This is also the honest limit of a ranked-seller tool like Farmrix: it tells you which owners are likely to sell, it does not price the house for you, so you still run the comps.

Why an automated estimate is not a comp

An automated valuation is a starting hint, and the firms that publish them admit as much in their own accuracy numbers. Redfin reports that its estimate carries a median error of 1.85 percent for listed homes and 7.26 percent for off-market homes, across 92 million properties, updated September 2025. Sit with that second figure. On a $500,000 house a 7.26 percent median error is about $36,300, and median means half of all homes miss by even more than that.

For a wholesaler working a $40,000 spread, a $36,000 swing in value is the entire deal. That is the gap between a model averaging across millions of homes and three sales you picked and adjusted yourself for the specific kitchen in the specific house. The estimate is fine for a first glance to see if a lead is worth an hour. It is not the number you sign a contract on, and it is a different tool from a ranked list of likely sellers, which answers who might sell rather than what the home is worth.

A worked example

Take a subject of 1,600 square feet, 3 bed, 2 bath, built in 1994, average condition, with a two-car garage. You pull three arms-length sales that closed in the last five months within the same subdivision. The dollar adjustments below are illustrative, the kind of market-based figures you would confirm from paired sales in your own zip, not fixed national values.

The raw sales ran from $299,000 to $325,000, a $26,000 spread wide enough to talk yourself into almost any number. After adjustment they cluster at $305,000, $309,000 and $311,000. You anchor around $310,000, and you can say exactly why, line by line. The highest raw sale would have handed you $325,000 and a bad offer.

Mistakes that blow up your number

Most bad comps come from the same short list of shortcuts. Each one has a fix that costs a few extra minutes.

• Using active listings as comps. An asking price is not a sale. In a slow market half of them will close under ask, and you have anchored high.
• Grabbing the three highest sales. Bracket instead. Without a comp below your subject you have no floor.
• Ignoring condition. The gutted 2024 remodel and the 1988 original are not the same house at the same price, whatever the square footage says.
• Trusting one automated estimate. A 7.26 percent median off-market error is real money on a five-figure margin.
• Mixing seasons in a moving market. A sale from a hot April can overstate a subject you are pricing in a slow November.
• Counting a non-arms-length transfer. A sale between relatives or a bank REO is not what a normal buyer would pay.

What to do next

Run your next number in five steps. Write the subject profile before you look at a price. Pull six to eight closed sales from the last 6 to 12 months in the same market area. Cut to the three or four that bracket the subject above and below. Adjust each one to the subject with figures you can defend. Then anchor to where the adjusted values cluster, not to the highest raw sale.

Comps tell you what a house is worth. They do not tell you which owner is ready to sell it, and that is a separate search. Farmrix scores every owner in a market on how likely they are to sell in the next 6 to 12 months and mails the top of that ranked list, so your comping hours go to houses that might actually trade rather than the whole county. Price the deal with comps, find the deal with the list. If you run fix-and-flips, do the honest version of this once: pull your next ten comps by hand, before any tool does it for you, and you will spot the bad ones on sight from then on. Our page for flippers shows where that fits in a deal.