The NVD (National Vulnerability Database) is the US government repository of vulnerability data, operated by the National Institute of Standards and Technology (NIST), that takes every published CVE record and adds severity scores, weakness classifications and affected-product data. In cybersecurity, and specifically in the management of vulnerabilities, the NVD is the most widely consumed public source for answering three questions about a known flaw: how severe it is, what kind of weakness it represents and which software versions it affects.
The NVD does not assign identifiers and does not discover flaws. It is an analysis and enrichment layer on top of the CVE Program, and its value lies in structured, machine-readable metadata that scanners, patch management tools and compliance frameworks consume automatically.
How the NVD enriches CVE records
The NVD works as a downstream consumer of the CVE List. When a CVE Numbering Authority (CNA) publishes a new record, the NVD ingests it within minutes, and NIST analysts later perform enrichment: they review the description and references, then attach standardized metadata that the original record may lack.
Enrichment adds three core data types. A CVSS base score expresses technical severity, a CWE identifier names the underlying weakness type, and CPE applicability statements specify exactly which vendors, products and version ranges are vulnerable. The following table shows what each layer contributes.
| Data element | Standard used | Question it answers |
|---|---|---|
| Severity score | CVSS v3.1 and v4.0 | How serious is the flaw in isolation |
| Weakness type | CWE | What kind of coding or design error caused it |
| Affected products | CPE 2.3 match criteria | Which products and versions are vulnerable |
| Reference tags | NVD reference taxonomy | Which links are patches, advisories or exploits |
| Analysis status | NVD status values | How far NIST has processed the record |
Each record also carries a lifecycle status. Typical values are Received, Awaiting Analysis, Undergoing Analysis, Analyzed, Modified, Deferred and Rejected. A record marked Awaiting Analysis has an identifier and description but may have no NIST score and no CPE data, which matters for any tool that relies on those fields.
Primary and secondary scores in the NVD
The NVD distinguishes between scores assigned by NIST and scores supplied by the CNA. NIST scores are labeled as primary, while scores provided by the vendor or another CNA appear as secondary. Both can be shown on the same record, and they frequently disagree.
Disagreement arises because NIST scores from the public description, usually under worst-case assumptions, while vendors know configuration defaults and mitigating factors. A vendor may rate a flaw 6.5 because exploitation needs a non-default setting, while the NVD rates it 9.8. Mature programs record which source they used and document the reason when they override a score.
| Score source | Typical strength | Typical weakness |
|---|---|---|
| NIST primary score | Consistent method across all vendors | Conservative, limited product knowledge |
| CNA secondary score | Informed by product internals | Possible bias toward lower severity |
| CISA ADP score | Fills gaps during NVD backlogs | Coverage varies by record |
The NVD stopped producing CVSS v2 scores for new records in 2022, and it added support for CVSS v4.0 data in 2024. Older records may therefore hold only v2 vectors, which are not directly comparable with v3.1 values.
How the NVD relates to CVE, CVSS, CWE and KEV
The NVD is often confused with the CVE Program, yet the two have different owners and roles. The CVE Program, run by MITRE with CISA sponsorship, assigns identifiers and publishes records; the NVD analyzes and enriches them. CVSS and CWE are standards the NVD applies, while CISA KEV and EPSS are separate signals that complement NVD severity.
| Resource | Operator | Main role |
|---|---|---|
| CVE List | CVE Program (MITRE, CNAs) | Unique identifiers and base descriptions |
| NVD | NIST | Scores, weaknesses and affected products |
| CVSS | FIRST | Scoring framework for technical severity |
| CWE | MITRE | Taxonomy of software and hardware weaknesses |
| CISA KEV | CISA | Catalog of flaws exploited in the wild |
| EPSS | FIRST | Probability of exploitation within 30 days |
The practical distinction is between severity and risk. The NVD tells a defender how damaging a flaw could be; KEV and EPSS indicate how likely it is to be exploited. Prioritizing by NVD scores alone tends to produce an unmanageable list of critical items.
Accessing NVD data in practice
Most organizations never read NVD pages directly; they consume the data through scanners, software composition analysis tools and asset inventories. These tools match installed software against CPE match criteria to decide whether a host or dependency is affected.
Direct access is through the NVD API 2.0, which replaced the legacy JSON data feeds that were retired in 2023. The API offers endpoints for CVE records, change history and the CPE dictionary, and it enforces rate limits: roughly 5 requests per rolling 30 seconds without a key and about 50 with a free API key.
| Need | Recommended method | Reason |
|---|---|---|
| Look up one flaw | NVD web interface | Human-readable view with history |
| Keep a local mirror | API with incremental sync | Fetch only records changed since last run |
| Feed a scanner | Vendor tool using NVD data | CPE matching handled by the tool |
| Audit a score change | CVE change history endpoint | Shows who changed what and when |
Teams building their own integration should sync incrementally using last-modified date ranges, cache results locally and respect the published rate limits, because aggressive polling leads to throttling and failed jobs.
Limitations and common mistakes
The NVD is authoritative but not complete or instantaneous. Since early 2024 NIST has faced a significant enrichment backlog, leaving many thousands of records without NIST scores or CPE data for months. CISA responded with its Vulnrichment program, which adds scores and CWE data as an authorized data publisher.
- Treating the base score as risk: a 9.8 score ignores whether the asset is exposed, whether exploits exist and what compensating controls apply.
- Assuming no score means low severity: an unscored record may simply be waiting in the backlog.
- Trusting CPE matches blindly: inaccurate or broad version ranges produce false positives and false negatives in scanners.
- Ignoring record changes: scores and applicability are revised, so a cached copy can become stale.
- Relying on a single source: vendor advisories, KEV and EPSS should complement NVD data.
A sound program uses the NVD as a baseline reference, verifies applicability against vendor advisories and layers exploitation intelligence on top before setting remediation deadlines.
In everyday language
Think of the CVE system as a library that gives every known software flaw its own catalog number. The NVD is the reference desk next to that catalog. Experts there read each entry and add notes: how dangerous the flaw is, what kind of mistake caused it and exactly which programs and versions have it.
Security tools read those notes automatically to tell companies which of their computers need fixing. The key point is simple: the NVD describes how bad a flaw could be, but it does not say whether criminals are actually using it, so careful teams check other sources too.
Seen in the real world
A security engineer at a regional hospital network runs a weekly vulnerability scan across about 4,000 servers and workstations. The scanner pulls CPE match data from the NVD and flags 1,200 findings, 310 of them rated critical by the NVD base score. Patching all of them within the hospital’s 15-day critical window is not realistic.
The engineer cross-checks the critical list against the CISA KEV catalog and EPSS probabilities, and also reviews vendor advisories where the NVD and the vendor disagree. Two findings turn out to be false positives caused by an over-broad version range, and one new flaw on an internet-facing VPN appliance is still Awaiting Analysis with no NIST score at all, yet appears in KEV.
The team patches the VPN appliance within 48 hours and schedules the remaining items by exposure. The outcome is a focused list of 40 urgent fixes instead of 310, with every decision traceable to NVD, KEV and vendor data.