FMEA and Fault Tree Analysis attack risk from opposite ends. An FMEA works bottom-up: list every way each component or process step could fail, then score each mode by Severity × Occurrence × Detection. A fault tree works top-down: start from one unacceptable outcome — a fire, a crash, a data loss — and decompose it through AND/OR logic gates into the combinations of basic events that could cause it. The fastest way to choose: are you screening many failure modes, or dissecting one catastrophic event? Screening → FMEA. Dissecting → FTA.
Side by side
| FMEA | Fault Tree Analysis | |
|---|---|---|
| What it is | Structured risk-assessment worksheet | Boolean logic model of one failure event |
| Direction | Bottom-up, inductive (mode → effect) | Top-down, deductive (event → causes) |
| Question it answers | "What could fail, and which modes matter most?" | "How exactly could this one disaster happen, and how likely is it?" |
| Scope | Broad — every failure mode in the system | Deep — one undesired top event at a time |
| Scoring | Ordinal: RPN / AP (1-10 scales) | Probabilistic: real failure rates, gate math, cut sets |
| Handles combinations? | Poorly — one mode at a time | Natively — AND gates model multi-cause failures |
| Output | Ranked worksheet + actions + residual risk | Tree diagram, minimal cut sets, top-event probability |
| Regulatory home | IATF 16949, AS9100, FDA design controls | NRC (nuclear), ARP4761, IEC 61508, ISO 26262 |
| Team & time | Cross-functional workshop, 2-20 h | Analyst-led with expert input, 4-40 h |
The key insight: breadth screens, depth dissects
FMEA's strength is coverage. Nothing else forces a team to walk through every component or process step and ask how it could fail — which is why regulators love it as the default risk tool. But its weakness is combinations: FMEA evaluates one failure mode at a time, so a disaster that requires two simultaneous failures (pump fails AND backup fails) slips between the rows.
That is exactly where the fault tree lives. An AND gate says "this happens only if all inputs happen" and multiplies probabilities accordingly; minimal cut sets expose which combinations are single points of failure in disguise. The price is scope — each tree models one top event, so you cannot fault-tree your whole product. The mature pattern in safety-critical work: FMEA to screen everything, then FTA on the handful of catastrophic top events the FMEA flagged.
Choose FMEA when…
- You need broad risk screening across a whole design or process — the classic pre-launch DFMEA/PFMEA.
- Your customer or auditor expects it. Automotive and aerospace supplier quality manuals name FMEA explicitly; a fault tree will not close that requirement.
- Failures are mostly independent and the worst credible outcome is costly rather than catastrophic — RPN/AP prioritisation is enough to direct prevention effort.
- You want a living document the team updates as the design evolves and lessons (e.g. from an 8D) feed back in.
Choose Fault Tree Analysis when…
- One specific outcome is unacceptable — loss of life, fire, release, total data loss — and you must understand every path to it.
- Failures interact. Redundant systems, backup chains, human-plus-machine sequences — AND/OR logic is built for exactly this.
- You need a number. Certification or safety cases that require "probability of top event ≤ 10⁻⁹ per flight hour" can only come from quantitative FTA, never from RPN.
- You are investigating a complex incident backward — FTA also works reactively, reconstructing how a real event could have happened.
The 30-second decision rule
Many modes, broad screen, auditor-friendly worksheet → FMEA. One catastrophic event, interacting failures, need a probability → FTA. Safety-critical program? Run both: FMEA finds the monsters, FTA dissects them. For everyday quality problems, neither may be the right first tool — a Pareto or fishbone is faster.
Score risks with the free FMEA tool →Two mistakes that waste weeks
Mistake 1: fault-treeing everything. FTA is expensive — hours per tree, expert-hungry, and each tree covers exactly one top event. Teams that try to model every failure this way burn weeks producing beautiful diagrams for events that a 2-hour FMEA pass would have deprioritised. Screen first; dissect only what earns it.
Mistake 2: trusting RPN where combinations rule. An FMEA row scores "backup pump fails" as low-risk (rare, detectable) and moves on. But if the primary pump's failure and the backup's failure share a cause — same contaminated fuel, same maintenance skip — the real probability of losing both is far higher than either row suggests. Common-cause and multi-event risks need gate logic, not ordinal scores.
Frequently asked questions
What is the main difference between FMEA and FTA?
Direction. FMEA is bottom-up and inductive (failure mode → effect, scored by S×O×D). FTA is top-down and deductive (one undesired event → the logic of causes behind it).
Can FMEA and FTA be used together?
Yes — the standard safety-critical pattern: FMEA screens all modes broadly, FTA models the few catastrophic top events in depth. ARP4761 and ISO 26262 expect both.
Which is better for regulatory compliance?
Depends on the standard: IATF 16949 / FDA expect FMEA; NRC, ARP4761, IEC 61508 and ISO 26262 additionally expect quantitative analysis like FTA for catastrophic events.
Is FTA quantitative or qualitative?
Both — the tree and its minimal cut sets are qualitative insight; assigning failure rates to basic events makes the top-event probability computable.
When is FMEA alone enough?
When no single failure is catastrophic and you mainly need broad screening with prioritised actions — most manufacturing processes and non-safety-critical products.
Related resources
- FMEA: the complete guide — RPN formula, rating scales, DFMEA vs PFMEA, regulatory context
- Fault Tree Analysis guide — AND/OR gates, minimal cut sets, probability calculation
- Fishbone vs Fault Tree — qualitative brainstorm vs the probability tree
- Fishbone vs FMEA — reactive cause-mapping vs proactive scoring
- All five RCA tools compared — the full decision guide