Kirkpatrick's Four Levels of Evaluation

6 min read · SessionData

In 1959, Donald Kirkpatrick completed his doctoral dissertation at the University of Wisconsin. In it, he proposed something deceptively simple: that training evaluation happens at four levels. Not one. Not a single score. Four distinct questions, each harder to answer than the last.

The framework spread through corporate training, government agencies and universities. It was refined, debated and occasionally misused — but never replaced. Every serious evaluation model since has either built on it or defined itself against it.

The four levels

Level 1 — Reaction

Did participants respond favourably to the training? This is the smile sheet, the post-session survey, the "how was it?" question. It measures satisfaction, relevance and engagement.

Reaction data is easy to collect and hard to ignore. A session that people disliked is a session that has a problem. But satisfaction alone proves nothing about whether learning occurred — enjoyable is not the same as effective.

Level 2 — Learning

Did participants acquire the intended knowledge, skills or attitudes? This level asks whether the content actually landed — whether something changed between the start of the session and the end.

Measurement here typically involves assessments: pre-and-post tests, skills demonstrations, scenario-based judgement tasks. It answers a different question from Level 1: not "did they like it?" but "did they get it?"

Level 3 — Behaviour

Are participants applying what they learned back on the job? This is where evaluation gets genuinely difficult. Knowledge that stays in the classroom has limited value. Level 3 asks whether the learning transferred — whether people are actually doing things differently.

Behaviour change takes time, depends on the work environment, and is influenced by factors well beyond the training itself. Managers, systems, incentives, culture — all of these can support or suppress transfer. This is why Argyris's double-loop learning matters here: sometimes the problem is not that people failed to learn, but that the system they returned to made the new behaviour impossible.

Level 4 — Results

Did the training produce the outcomes the organisation was after? Lower error rates, faster onboarding, higher retention, improved safety records — the business metrics that justified the investment in the first place.

Level 4 is the hardest to isolate. Training is rarely the only variable. Attribution requires careful design — control groups, longitudinal tracking, honest accounting of what else changed at the same time. But it is the level that speaks the language of the people who fund the work.

The chain, not the menu

A common misreading of Kirkpatrick is to treat the levels as a menu — pick the one that suits your budget. In practice, they form a chain. Positive reactions do not guarantee learning. Learning does not guarantee behaviour change. Behaviour change does not guarantee results.

Each level is necessary but not sufficient. Skip a level and you cannot explain what happened at the next. A programme that produced good results but measured nothing in between has a story it cannot tell — and cannot reliably repeat.

What Kirkpatrick did not claim

Kirkpatrick was careful to frame his model as a taxonomy of outcomes, not a causal ladder. He did not claim that satisfaction causes learning or that learning causes behaviour change. Each level describes a different kind of evidence. The chain is inferential, not mechanical.

Later practitioners — particularly Jack Phillips with his Level 5 ROI extension — pushed the framework toward financial proof. This produced useful discipline but also a persistent confusion: the belief that unless you can calculate a dollar return, evaluation has failed. Kirkpatrick himself resisted this. The purpose of evaluation, he argued, was to improve programmes — not to justify them after the fact.

The gap the model reveals

The enduring power of the four levels is not prescription but diagnosis. Most organisations evaluate at Level 1. Some reach Level 2. Very few sustain Level 3 or Level 4.

The gap is not for lack of wanting. It is structural. Reaction surveys are cheap and immediate. Behaviour observation requires time, access and a baseline. Results attribution requires design decisions made before the programme runs — not after.

This is why frameworks like Kolb's learning cycle and Senge's learning organisation matter alongside Kirkpatrick. They describe the conditions under which learning transfers and sticks — the upstream factors that determine whether Level 3 and Level 4 are even possible.

Kirkpatrick, D. L. (1959). Techniques for evaluating training programs. Journal of the American Society of Training Directors, 13(11), 3–9.

Kirkpatrick, D. L., & Kirkpatrick, J. D. (2006). Evaluating Training Programs: The Four Levels. 3rd ed. Berrett-Koehler.