510(k) Predicate Selection Isn't a Similarity Contest — It's a Two-Prong Legal Test

Most 510(k) predicate selection happens the same way: someone searches FDA's 510(k) database, finds the closest cleared device, and cites it. That instinct is understandable and, on its own, insufficient. The legal test in 21 CFR 807.100 does not ask whether a device looks similar — it asks two narrower questions, and a predicate that fails either one does not just weaken your submission. It restarts the clock on the whole 510(k) clearance effort.

The test is narrower than "looks similar"

Section 513(i) of the FD&C Act and 21 CFR 807.100 set the actual standard. First prong: the new device must have the same intended use as the predicate — not an overlapping one, not a broader one, the same one. Second prong: the new device must have either the same technological characteristics as the predicate, or different technological characteristics that do not raise different questions of safety and effectiveness, supported by data — including clinical data where the Commissioner deems it necessary — showing the new device is at least as safe and effective. A device that is 90% similar to a predicate but fails either prong outright is not substantially equivalent; a device that differs meaningfully in materials or energy source can still clear if the data closes that gap. Resemblance is not the test. The two prongs are.

The two failure patterns reviewers actually see

The first is the split predicate: citing one device to satisfy the intended-use prong and an unrelated device with a different intended use to satisfy the technological-characteristics prong. It reads as substantial equivalence on paper while never actually comparing one device to itself on both prongs at once, which is why FDA's 2014 guidance closed it off. Citing multiple predicates is still allowed — but every predicate in the comparison has to share the subject device's intended use; the practice supports a coherent equivalence argument, not a patchwork of convenient partial matches. The second pattern is predicate creep: a chain of 510(k)s where each new device cites the prior clearance as its predicate, and small, individually defensible changes compound over several cycles into a device whose risk profile has quietly drifted from where the chain started. A predicate that is itself several generations removed from a directly comparable device deserves scrutiny before you inherit whatever accumulated in between, and the underlying design history file is where that lineage actually gets traced.

• Run both prongs explicitly, in writing. Document the intended-use match and the technological-characteristics comparison as two separate findings, not one impression of overall similarity.
• Trace the predicate's own ancestry. If your predicate was itself cleared against an earlier predicate, pull that chain and check where the technological characteristics started diverging from the device you actually intend to compare against.
• Check the predicate's safety record. Recalls, MAUDE reports, and design-related adverse events tied to your candidate predicate become your problem the moment you cite it.
• Never split the comparison across two devices. One predicate (or a coherent set that share intended use) has to carry both prongs of the test.

A predicate that clears the database search is not the same thing as a predicate that clears the two-prong test. Confirming the difference before you file is cheaper than discovering it in an Additional Information request. Why predicate selection is a legal determination, not a search result

What FDA's 2023 draft adds, and why it isn't really optional

In September 2023, FDA issued a draft guidance, Best Practices for Selecting a Predicate Device to Support a Premarket Notification [510(k)] Submission, that layers four quality screens onto the legal test: was the predicate cleared using well-established methods (recognized consensus standards, FDA guidance, or accepted published methods); does it meet or exceed expected safety and performance; is it free of unmitigated use-related or design-related safety issues; and is it clear of an associated design-related recall. As of this writing the guidance remains in draft — it has not been finalized, and its recommendations are explicitly non-binding. That does not make the underlying question irrelevant. Reviewers who have read the draft are already asking these questions informally, and the guidance recommends submitters include a narrative in the 510(k) Summary explaining how the predicate was chosen and how these considerations factored in. Writing that narrative before a reviewer asks for it — and grounding it in the same risk management file that already documents the device's safety profile — costs little and removes a predictable source of review-cycle delay.

None of this changes the shape of the 510(k) pathway — it changes how defensible the first decision in that pathway is. Teams that treat predicate selection as a database search inherit whatever risk, drift, or safety history came with the device they picked. Teams that treat it as the two-prong legal determination it actually is spend less time answering Additional Information requests about a choice that should have been settled before submission. If a predicate candidate looks close but not exact, that is exactly the case a 510(k) clearance strategy engagement or a Pre-Submission conversation exists to resolve.