How credit scoring works

A U.S. consumer's credit score is, in mathematical terms, a regression model's prediction of the probability that the consumer will become 90+ days delinquent on a credit obligation in the next 24 months, scaled to a familiar 300-to-850 range. The model is trained on millions of historical consumer credit-bureau records, and its inputs are the contents of the consumer's file at one of the three nationwide credit bureaus. The score is then sold by the scoring company (Fair Isaac Corporation for FICO scores, the VantageScore joint venture for VantageScore) to lenders, who use it as one input — typically the most important single input — in deciding whether to extend credit and on what terms.

This article describes the credit-bureau ecosystem, the two principal score families, the five factor groups in a typical FICO score, and what is and is not included in the calculation. Consumer-protection rights around credit reporting are governed by the Fair Credit Reporting Act and Regulation V; the dispute mechanics are part of the same framework.

The three bureaus

The "nationwide consumer reporting agencies," as defined by the Fair Credit Reporting Act, are Equifax, Experian, and TransUnion. Each maintains a separate consumer file on most U.S. adults, containing identification information, account history (each open and closed credit account, with payment history, balance, and limit), public records (bankruptcies; previously also tax liens and civil judgments, which were largely removed in 2017–2018 under the National Consumer Assistance Plan), and inquiry history (each "hard pull" and certain other inquiries).

The three bureaus are private companies that compete for furnisher (lender) and consumer business. They are not coordinated; data furnished to one bureau is not automatically furnished to the others. As a result, a consumer's file at one bureau can differ meaningfully from the same consumer's file at another: a creditor may report to only one or two bureaus rather than all three, a public-record event may appear at different times, errors are corrected independently at each bureau. Credit scores calculated from each bureau's file will typically be close but not identical for the same consumer.

The FCRA gives consumers a right to a free annual report from each bureau, which can be obtained through AnnualCreditReport.com (the only website authorized for this purpose under the FCRA). In response to the COVID-19 pandemic, the bureaus expanded free-report availability to weekly and have committed to maintaining the weekly cadence permanently; verify against the bureaus' current policies before relying on a specific cadence.

FICO and VantageScore

Two principal score families are in widespread use in U.S. consumer credit.

FICO scores are produced by Fair Isaac Corporation. The FICO family has multiple versions in active use — FICO 8, FICO 9, and FICO 10/10T are the most-cited consumer versions — and FICO produces industry-specific score variants for auto lending, mortgage lending (the older FICO 2, 4, and 5 variants are still the standard for U.S. residential mortgage underwriting through Fannie Mae and Freddie Mac), and bankcard underwriting. All FICO scores use the 300–850 scale; higher scores indicate lower predicted default risk. FICO has been the dominant U.S. credit-score family since the 1990s.

VantageScore is a competing score family produced by VantageScore Solutions, a joint venture of Equifax, Experian, and TransUnion. VantageScore 3.0 and 4.0 are the principal versions in current use; the 300–850 scale was aligned with FICO with version 3.0 (earlier versions used different scales). VantageScore has gained share in non-mortgage consumer lending and is the score family used by many free-credit-monitoring services. The Federal Housing Finance Agency in 2022 designated VantageScore 4.0 as approved for use in mortgage underwriting at Fannie Mae and Freddie Mac alongside the FICO classic models; the transition implementation has been the subject of ongoing rulemaking and may affect mortgage scoring practices over the next several years.

The two families are correlated but not interchangeable. A consumer's FICO and VantageScore numbers will typically be within 20 to 50 points of each other but can diverge more substantially in specific cases (thin files, recent score changes, or specific account patterns the two models weight differently). Lenders' published rate sheets cite specific score families and versions; a consumer comparing offers should match the score type to the rate quoted.

The factor groups

FICO publicly disclaims a single algorithm, but it discloses approximate weights for the five factor groups that drive a typical FICO 8 score:

• Payment history (~35%). Whether and how late each credit account has been paid, weighted by recency and severity (a 30-day late payment has less effect than a 60-day, which has less effect than a charge-off or bankruptcy). This is the single most important factor and the one most directly under the consumer's control.
• Amounts owed (~30%). Total credit utilization (balances divided by limits across all revolving accounts), the number of accounts with balances, and the proportion of installment-loan balances remaining. Utilization is the most-cited single sub-factor here; the conventional wisdom that "below 30% utilization" is good is approximate, with lower utilization generally producing higher scores.
• Length of credit history (~15%). The age of the consumer's oldest credit account, the average age across all accounts, and how recently each account was used. Closing an old account can shorten the average age and (with a delay) reduce the score; opening many new accounts can have the same effect.
• New credit (~10%). The number of recent hard-pull inquiries and recently opened accounts. Each hard pull typically has a small score effect (often less than 5 points) that fades within 12 months; multiple inquiries within a short window for the same product category (mortgage, auto, student loan) are typically grouped as a single inquiry under FICO's "rate-shopping window" — the window length differs by score version but is typically 14 to 45 days.
• Credit mix (~10%). The variety of credit account types (revolving, installment, mortgage). The effect is typically small and is at the margin of the score.

VantageScore's published weight structure differs in detail but addresses similar factors: payment history is the largest weight, followed by total credit usage, then length and types of credit, with smaller weights for recent inquiries and available credit. The relative weights vary by version.

What is not in a credit score

U.S. credit scoring excludes information that is either not in the credit-bureau file or that the score model is designed not to consider. The exclusions matter both substantively and as a matter of law:

• Income, employment, and assets: not part of the credit-bureau file (except as a generally outdated indicator the bureau may carry for some consumers); not used in credit scoring. Lenders collect this information separately for underwriting but it does not enter the score.
• Race, color, religion, national origin, sex, marital status, and age: prohibited as scoring inputs under the Equal Credit Opportunity Act and Regulation B. The bureaus do not collect this information.
• Soft pulls (credit-monitoring inquiries, prequalification inquiries, account-review inquiries by existing creditors): visible only to the consumer on their bureau-direct report and do not affect the score.
• Bank deposit balances, transaction history, savings rates: not in the credit bureau file. Specialty agencies like ChexSystems and Early Warning Services maintain separate deposit-account histories, used for new-account screening but not for credit scoring. See ChexSystems and being denied an account.
• Utility, rent, and telecommunications payment history (in most cases): historically not reported to the credit bureaus and not in the score. Some recent products allow consumers to opt into reporting of rent and utility payments through services like Experian Boost, with limited effect on scoring; the broader effort to include "alternative data" in mainstream scoring has progressed gradually.

Hard pulls, soft pulls, and dispute

A "hard pull" (hard inquiry) is a credit-bureau inquiry made as part of a credit application. Hard pulls appear on the consumer's credit-bureau file and are visible to other lenders. Each hard pull typically reduces the score by a small amount that decays over time; multiple hard pulls within a rate-shopping window for the same product category are typically treated as a single inquiry under FICO and VantageScore models. A "soft pull" (soft inquiry) — prequalification checks, account-review by existing creditors, the consumer's own self-pull — does not appear on the credit-bureau file in a way that affects the score and is not visible to other lenders.

If a consumer believes their credit report contains inaccurate information, the FCRA gives them the right to dispute it directly with the bureau, the furnisher, or both. The bureau must investigate within 30 days (sometimes extended to 45) and either correct, delete, or substantiate the disputed item. The CFPB has identified credit-reporting accuracy as a recurring supervisory and enforcement priority; cumulative remediation through CFPB actions has produced large adjustments to bureau records over the past decade.

Limits and uncertainty

The credit-scoring framework as described here is durable; specific FICO and VantageScore version weights and methodology details are proprietary and change with new versions. The transition to VantageScore 4.0 and the latest FICO models in mortgage underwriting is ongoing under FHFA oversight and may change which model affects a given mortgage borrower. The CFPB's continued attention to credit-reporting accuracy, the FCRA's evolving treatment of medical-debt and other specific information categories, and the gradual movement toward "alternative data" inclusion in scoring all suggest evolution at the edges of the system. The core model — predictive scoring trained on credit-bureau data, with the five-factor structure roughly as described — is unlikely to be replaced in the foreseeable future.