Two theories of performance in multiple‐choice tests, one based on finite states and the other based on continuous distributions, are compared in this paper as regards their capability to account for data from an answer‐until‐correct administration of a test of spatial reasoning. The theories are shown to be almost equivalent in their predictions and to fit the data almost equally well, with a minimal advantage in favour of the continuous distribution theory. A theoretical analysis of the models shows, though, that the finite state theory is closer than the continuous distribution theory to an accurate representation of the processes involved in responding to test items. The finite state theory is also shown to be more general and flexible than the continuous distribution