Situation Awareness — The Endsley Three-Level Model and Its Critics

emerging Updated: 2026-06-13
situation-awareness endsley-model three-level-model mental-models working-memory sensemaking distributed-situation-awareness team-sa comprehension projection

Situation Awareness — The Endsley Three-Level Model and Its Critics

Status: emerging
Last updated: 2026-06-13
Sources: Situation Awareness Misconceptions And Misunderstandings Endsley 2015.Pdf
Tags: [situation-awareness, endsley-model, three-level-model, mental-models, working-memory, sensemaking, distributed-situation-awareness, team-sa, comprehension, projection]

Summary

Situation awareness (SA) — knowing what is going on around you well enough to act — is one of the most cited constructs in human factors, and Endsley's (1995) three-level model of perception, comprehension and projection is its most influential account. Endsley (2015) answers a wave of recent critiques of that model, arguing that seven of the most common objections rest on misreadings of what the model actually claims, and then compares the model against three proposed replacements: Situated SA, Distributed SA, and the Data-Frame model of sensemaking. The article matters to this knowledge base because SA is the theoretical target that gaze behaviour is so often used to measure: the model's first level — perception of relevant elements — is precisely what eye-tracking indexes, and its goal-driven account of attention explains why operators look where they do. It sets out what SA is, what it is not, and where genuine disagreements remain.

Body

Context

Endsley (2015), writing in a Journal of Cognitive Engineering and Decision Making special issue, reviews and rebuts criticisms of the Endsley (1995) three-level model of SA and weighs newer rival models against it (PDF p. 1, orig. p. 4). The lens is theoretical defence and comparison rather than new data: the paper restates the model, then takes seven recurring criticisms in turn as "fallacies" born of misunderstanding, and finally examines Situated SA, Distributed SA, and sensemaking. Within this knowledge base it is the situational-awareness anchor — the construct that the KB's gaze-as-indicator strand ultimately measures — and it connects to the attention-allocation work in Expected Value Model Of Visual Scanning and Attention Allocation Driver Model (both formalise the goal- and value-driven attention that, in Endsley's model, feeds Level 1 perception) and to the working-memory account in Pupil Dilation Cognitive Load.

Key Points

The model has three ascending levels plus goal-driven processing. SA in the Endsley (1995) model comprises perception of relevant elements (Level 1), comprehension of their meaning (Level 2), and projection of future states (Level 3); it also specifies the roles of goals, mental models, schemata, expectations, attention, and working memory in forming and maintaining that awareness (PDF pp. 2–4, orig. pp. 5–7). Endsley stresses that the model is a framework built on well-researched cognitive constructs, and that its breadth — often raised as a weakness — is what lets it generate concrete design and training guidance (PDF p. 4, orig. p. 7).

Most criticisms misread the model (seven fallacies). Endsley addresses each: (1) the three levels are not strictly linear stages — they are ascending levels of awareness, and higher levels routinely drive the search for and default-filling of Level 1 data, so processing is iterative, not a one-way 1-2-3 flow (PDF pp. 5–6, orig. pp. 8–9); (2) the model is not a linear data-driven information-processing model — it explicitly couples data-driven and goal-driven processing (PDF p. 6, orig. p. 9); (3) the product-versus-process distinction critics reject is in fact maintained in the model (SA as a state vs. situation assessment as the process producing it); (4) the model is cyclical and dynamic, with feedback from comprehension and projection back to perception; (5) the model does account for meaning — comprehension (Level 2) is exactly the meaningful integration of elements relative to goals; (6) SA is not held entirely in working memory — experts rely on long-term memory structures, mental models and pattern-matching that circumvent working-memory limits (PDF pp. 10–12, orig. pp. 13–15); and (7) the model is not merely an "in-the-head" view divorced from the world — it gives environmental, task, and system factors, including interface design, a central role (PDF pp. 12–13, orig. pp. 15–16).

The rival models add less than they claim. Situated SA (Chiappe, Strybel & Vu, 2011) argues SA is distributed between the head and the environment to reduce working-memory load; Endsley replies that the mental-model and long-term-memory mechanisms in her model already explain low working-memory demand in experts, so a separate construct is unnecessary (PDF pp. 13–14, orig. pp. 16–17). Sensemaking (the Klein et al. Data-Frame model) is, on Endsley's reading, a backward-looking, effortful, conscious subset of SA processes — useful for novel problems and diagnosis — whereas SA also covers the rapid, automatic pattern-matching that experts use most of the time, so sensemaking is narrower than SA, not a replacement (PDF pp. 15–18, orig. pp. 18–21). Distributed SA (DSA) (Salmon, Stanton & Young, 2012; Stanton et al., 2010), which locates SA in the whole socio-technical system and treats agents as "owning part of" the SA, draws the sharpest disagreement: Endsley argues DSA gives no account of how information should be presented to build SA, and ignores 60 years of evidence that how information is displayed is critical to whether people can use it (PDF p. 25, orig. p. 28).

Conclusion

Endsley (2015) concludes that the construct of SA has moved "from revolutionary newcomer to the mainstream", that the recent critiques rest largely on misconceptions of the 1995 model which she has corrected here, and that the model's value lies in the roughly 50 design guidelines and the SA-measurement methods (e.g. SAGAT) it has generated for individuals and teams (PDF pp. 25–26, orig. pp. 28–29). She allows that genuine disagreements will persist and hopes they can be resolved through objective research and accurate understanding of where the models actually differ. For this knowledge base the takeaway is a clear, defended definition of SA and its three levels, and an explicit warning relevant to gaze-based inference — Level 1 SA is perception of relevant elements, so direction of gaze indexes the data sampled, not yet the comprehension or projection that constitute higher SA.

References

Chiappe, D.L., Strybel, T.Z. and Vu, K.P.L. (2011) 'Mechanisms for the acquisition of situation awareness in situated agents', Theoretical Issues in Ergonomics Science, 13(6), pp. 625–647. To be validated.

Dekker, S. and Hollnagel, E. (2004) 'Human factors and folk models', Cognition, Technology and Work, 6, pp. 79–86. To be validated.

Endsley, M.R. (1995) 'Toward a theory of situation awareness in dynamic systems', Human Factors, 37(1), pp. 32–64. To be validated.

Endsley, M.R. (2004) 'Situation awareness: Progress and directions', in Banbury, S. and Tremblay, S. (eds.) A cognitive approach to situation awareness: Theory, measurement and application. Aldershot: Ashgate, pp. 317–341. To be validated.

Endsley, M.R. (2015) 'Situation Awareness Misconceptions and Misunderstandings', Journal of Cognitive Engineering and Decision Making, 9(1), pp. 4–32. doi: 10.1177/1555343415572631. endsley2015misconceptions

Klein, G. et al. A data-frame theory of sensemaking (the Data-Frame model of sensemaking). To be validated.

Salmon, P.M., Stanton, N.A. and Young, K.L. (2012) — distributed situation awareness critique. To be validated.

Stanton, N.A. et al. (2010) — distributed situation awareness. To be validated.

Open Questions

  • Endsley argues direction of gaze indexes Level 1 perception but not comprehension or projection. How far can eye-tracking measures be pushed toward indexing Level 2/3 SA, and where does the gaze–attention–comprehension chain break down? (Connects to the gaze-vs-attention caveat flagged for Proctor & Proctor in QUESTIONS.)
  • The disagreement with Distributed SA turns on whether SA is a cognitive state in individuals or a property of the socio-technical system. The corpus has no source presenting the DSA case in its own terms — only Endsley's rebuttal — so the tension is recorded one-sidedly.
  • SAGAT and related freeze-probe measures are referenced but not held in the corpus. Ingesting a primary SA-measurement source would let the KB connect this theory to its measurement strand.