Zillow Says $412K. Redfin Says $377K. Which Number Do You Trust?
Why AVMs disagree on the same house — and why the disagreement is the most useful number on the page.
Pull up the same three-bedroom on Zillow and Redfin and you'll routinely see two different numbers — sometimes $35,000 apart on a $400K house. Neither model is broken. Each is doing exactly what it was built to do, on different data, with different assumptions. The mistake is treating either number as the value.
If you buy houses for a living, the disagreement between automated valuation models (AVMs) isn't noise to average away. It's the most useful signal you can get before you spend real money on diligence. Here's how to read it.
Why do Zillow and Redfin show different values for the same house?
Zillow and Redfin run separate automated valuation models on different data. They select different comparable sales, weight property facts differently, refresh on different schedules, and tune their models toward different goals. Two well-built models fed different inputs will produce different outputs — on the same house, on the same day.
The specific mechanics that drive divergence:
- Comp selection. Each model draws its own set of comparable sales — different search radius, different lookback window, different similarity scoring. Change the comps and you change the answer.
- Data inputs. Zillow blends public records, MLS feeds, and user-submitted edits (homeowners can change their own property facts). Redfin, as a brokerage, leans heavily on direct MLS data where it operates. Same house, different fact sets.
- Refresh cadence. One model may have ingested last week's neighborhood sale; the other hasn't yet. In a fast-moving zip code, a few weeks of lag is a real dollar difference.
- Model objective. A consumer portal optimizes for engagement across 100 million homes. A lender-grade AVM optimizes for defensibility on the subset it's confident about. Those are different jobs, and they produce different numbers.
And here's the structural part investors should internalize: the portals will never show you each other's numbers. Zillow won't display Redfin's estimate next to the Zestimate. Realtor.com's RealEstimate shows three AVMs (CoreLogic, Quantarium, and Collateral Analytics) — but not Zillow's or Redfin's. Every portal shows you one slice of the board and presents it as the picture.
How wrong can a single AVM estimate be?
Zillow's own published accuracy data puts the Zestimate's median error at roughly 2% for on-market homes but roughly 7% for off-market homes — and by definition, half of all errors are larger than the median. On a $300,000 off-market target, a 7% median miss is a $21,000 coin flip before you've accounted for condition.
(Zillow updates these figures over time; check their current published accuracy page for the latest. The on-market/off-market gap has been consistent for years.)
The off-market number is the one that matters for investors, because off-market is where acquisition teams and wholesalers actually operate. And it's the situation where every AVM is data-starved: no fresh listing photos, no agent-verified facts, public records that may be years stale. The model isn't wrong because it's badly built — it's wrong because it's guessing about a house nobody has described to it recently.
No AVM vendor has solved off-market. Which is exactly why the fix isn't "find a better single model."
What does it mean when AVM estimates are far apart?
A wide spread between AVMs usually means the property is hard to value algorithmically: sparse or mixed comps, unusual characteristics, conflicting public-record data, or a market in transition. Tight agreement means the house is "model-readable" — a standard property in a liquid market. Either way, the spread tells you how much to trust any single number, and what to check next.
A practical diagnostic:
- Tight cluster (sources within ~3–5%). The property is legible to algorithms. Comps are dense, records are clean. You can move faster with more confidence.
- One big outlier, rest clustered. Before dismissing the outlier, check that source's property facts. Wrong square footage, wrong bed count, or a bad parcel match happens constantly — and sometimes the outlier is the one that caught a real recent sale the others missed.
- Uniform scatter (everything disagrees with everything). Thin comps or a genuinely unusual property — acreage, mixed-use, heavy renovation, odd lot. This is a "walk it or pass" signal, not a "pick the middle" signal.
- Systematic split. When consumer portals cluster high and lender-grade models cluster low (or vice versa), you may be looking at a market in transition — one family of models has ingested the turn, the other hasn't.
The spread isn't a flaw in the data. The spread is the diligence to-do list.
How does consensus valuation work?
Consensus valuation pulls estimates from multiple independent AVMs and computes a central value — typically a median or confidence-weighted average — along with a band showing the full spread. The median resists a single bad source; the band tells you the size of your uncertainty in actual dollars.
This is not a new idea. It's what institutional acquisition desks have done manually for years: pull five to nine sources, look for agreement. Estimates clustered within about 3% of each other meant high confidence. A 15% divergence meant deeper diligence — or a pass. What's changed is that you no longer need a team and a stack of data subscriptions to do it.
A worked example
The numbers below are illustrative — a composite of the pattern we see constantly, not a specific live lookup:
| Source | Estimate |
|---|---|
| Zillow Zestimate | $412,000 |
| CoreLogic (via Realtor.com) | $405,200 |
| Collateral Analytics | $398,500 |
| Quantarium | $391,800 |
| Xome | $388,000 |
| Movoto | $379,900 |
| Redfin Estimate | $377,400 |
Consensus (median): $391,800. Spread: $34,600 — about 8.8% of the consensus value.
Read it like an operator: the middle of the board sits just under $392K, with a high-side outlier at $412K. If your deal only pencils at the Zestimate, the board is telling you the deal depends on the most optimistic model on the page. Check Zillow's property facts for that address before you believe it. If the deal pencils at $377K, you have margin the board supports. Either way, you learned more from seven numbers in one view than from any one of them alone.
How do investors put this into practice?
- Pull multiple sources on every target — before you offer, not after. At minimum, Zillow, Redfin, and Realtor.com's three-model RealEstimate are all free to check by hand.
- Log the spread, not just the middle. A $390K consensus with a $10K band and a $390K consensus with a $60K band are completely different deals.
- Diagnose outliers instead of deleting them. An outlier is either a data error you should know about or information the other models missed. Both matter.
- Adjust for condition yourself. No AVM can see deferred maintenance, a gut renovation, or the roof's age. Condition is your edge over every model.
- Overlay geo-risk. Flood zones, rail lines, transmission corridors, and highway proximity move real transaction prices, and AVMs largely don't price them. We've written about flood zone costs and transmission lines separately.
One lookup pulls up to 10 AVM sources — including the estimates the portals won't show side-by-side — then computes the consensus, the spread, and a geo-risk overlay. You're only charged when 3+ AVM sources respond (2+ with geo); otherwise your tokens auto-refund and you never pay for a thin result. See a sample report.
Frequently asked questions
Which is more accurate, Zillow or Redfin?
Neither wins everywhere. Each model publishes its own accuracy figures, measured on its own coverage with its own methodology, so the numbers aren't directly comparable. Accuracy varies by market, property type, and — most of all — whether the home is on-market or off-market. For investors, the more useful question is whether the two models agree on a specific property.
Why is my Zillow estimate so different from Redfin's?
Different comps, different data inputs, different refresh timing, different model objectives. A large gap usually means the property is hard to value algorithmically — sparse comps, conflicting records, or unusual characteristics — which is precisely the situation where you should trust no single number.
Should I just average Zillow and Redfin?
An average of two sources beats one source, but it's fragile: one bad input drags the average with it. A median across many independent sources resists outliers, and the spread between sources tells you more than any central number does. Wide spread = low confidence = more diligence before you offer.
Can an AVM replace an appraisal?
No. AVMs are automated estimates, not appraisals, and nothing here is a substitute for a licensed appraisal or your own diligence. Lenders require appraisals for a reason. AVMs — single or aggregated — are screening tools that tell you where to spend your diligence time and money.
AVM values are estimates, not appraisals, and should not be relied on as the sole basis for any transaction.
Run one address through the full board
Run your own spread. One address, up to 10 AVMs, consensus + geo-risk. 3 free lookups with a free account — no card.
Try AVMLens Free