Although the concept of risk has affected measurement methods for a long time, it seems not to be used generally in the evaluation of mathematics learning. In this paper we present some consequences of its use, and a particular kind of questionnaire which we call a loto-questionnaire. This appears as a cartesian product I x C of a set I, the set of items, and of set C, the set of choices. The set of items is a representative sample of possible questions concerning a mathematical subject, and the set of choices is derived from the learning requisites, so that the pattern of the correct answers may be statistically significant. After a short theoretical analysis of the constraints surrounding such questions, we present some results arising from their use. We think that loto-questionnaires may be useful for self-evaluation, but so far we have not used them for that purpose.