Contrary to traditional intelligence tests, dynamic measures of learning capacity have shown to provide reliable measures of children's general intellectual abilities and prove to be good predictors of future learning. In the present study we used a computerised version of the Hessels Analogical Reasoning Test (HART) to evaluate the changes in problem solving behaviour of children with and without learning difficulties as a result of training. Such training proves to be necessary as many children do not understand what is expected from them in such tasks and, as a consequence, do not use analogical reasoning to solve them. This affects the construct validity of the measure. The training focuses on the rules and procedures one needs to apply to be able to solve analogical problems, for example, systematic inspection of the matrix and the response alternatives, comparison of the different elements in the matrix and the inference and application of the relations found. In learning tests, it is generally assumed that children learn to engage in the processes needed for analogical problem solving during the training, and that the intra-individual variability in effective use of these processes at post-test is indicative of children's learning capacity. In this study we show that the training indeed provokes children to engage in the appropriate problem solving processes by analysing their visual behaviour. The data show that the children show more structured inspection patterns, more ‘intelligent' comparisons and spend more time on encoding the information in the matrix, which confirms the HART's construct validity after training.