For women who fail on the Piagetian water-level task, visual inspection and/ or verbal provision training procedures have either had no effect or led to only low to moderate improvement. The present experiment involved training within the observational learning paradigm and assessed its impact through multiple measurement. College women who failed a water-level item were selected as subjects. Pretested with different containers, they also supplied justifications to their answers and rated the certainty they associated with them. In the cognitive modelling group, a model performed part of the pretest series of horizontal water-line problems. In each first problem, she also verbalised her strategy, stressing the horizontality of the water surface and its parallelism with the surface on which the container rested. In the exemplary modelling group, the model acted identically, without voicing her strategy. In the no-modelling group, experimenter and subject chatted informally. There were two post-testing sessions, one followed immediately and the other 15 days later. Acquisition in the modelled series, as well as proximal generalisation in non-modelled but comparable series, occurred and persisted in modelling groups. In addition, enduring distal generalisation to non-modelled verticality was found in both groups, although it was higher in the exemplary modelling group. A similar pattern was obtained in justification production. However, positive correlations between post-test performance and certainty rating were infrequent. It appeared that exposure to modelled enactment of the rule-governed horizontality responses was the critical training component that fostered genuine improvement.