165 students, from the fifth, seventh and ninth grades were asked to solve written problems involving fractions. In the problems a reference quantity (RQ) was multiplied by a fraction (FR), yielding the value of a compared quantity (CQ). The fraction expressed either an part-whole ratio (PW), or a part-part ratio (PP). Six type of problems were made: the subject had to find a numerical value (RQ, FR or CQ) in one of the two ratios (PW or PP). Each problem was presented with three different settings (restaurant, woods, cake). Results show that at all ages the task CQ is better performed than FR, which is better performed than RQ. PW is better performed than PP and the effect of problem setting is negligible. For a given type of problem a same subject re-uses preferentially the same resolution procedures even when the subject can not solve the problems. The interpretation of the data suggest that the subjects implicitly categorize the six types of problems during the resolution process. The categories of problems are mastered in steps to a partially ordered hierarchy with multiple developmental paths. (Contains 1 table and 4 figures.)