In a longitudinal study of three levels of training evaluation, differentiated measures of trainees' reactions were shown to be more closely associated with learning outcomes than has been found with conventional reaction measures. However, reactions were generally unrelated to subsequent job behaviour. Both immediate and delayed learning were predicted by trainees' motivation, confidence and use of certain learning strategies and changes in job behaviour were independently predicted by transfer climate and learning confidence. It is preferable to measure training outcomes in terms of change from pre-test to post-test, rather than merely through attainment (post-test only) scores; the predictors of training outcomes were shown to differ according to which indicator is used. External factors better predicted learning changes (compared to predictions of post-test attainment) when their correlations with pre-test values differed substantially from their correlations with post-test scores.