Although there is a strong theoretical basis for concluding that conceptual models are effective in aiding users build mental models of computer systems, very little empirical evidence exists to support such a conclusion. Frequently, the effect has been weak. Subjects trained with a conceptual model often perform better than control group subjects in learning tests, but seldom at a statistically significant level. One possible reason for this is the influence of individual differences such as basic cognitive abilities. This study examined the influence of two cognitive variables - visual ability and learning mode - in the mental model formation process of novice users of an electronic mail filing system. We compared the effectiveness of two types of conceptual models-analogical and abstract. It was found that high-visual subjects performed significantly better than low-visual subjects. Abstract learners also performed better than concrete learners. More important, interaction effects were observed. Low-visual subjects were severely hampered by abstract models but performed as well as high-visual subjects when provided with analogical models. Abstract learners benefited from the abstract model but were hampered by the analogical model. On the other hand, concrete learners performed better with the analogical models compared to abstract models. The findings indicate that there is a need to consider individual differences, such as visual ability and learning mode, in research on mental models and on human-computer interaction in general.