Several methods to study the recognition and similarity of alphanumeric characters are briefly discussed and evaluated. In particular, the application of the choice-model (Luce, 1959, 1963) to recognition of letters is criticized. A feature analytic model for recognition of alphanumeric characters based on Tversky's (1977) features of similarity is proposed and tested. It is argued that the proposed model: (a) is parsimonious in that it utilizes a relatively small number of parameters, (b) is psychologically more meaningful compared with other approaches in that it is attempting to study underlying processes rather than just reveal a similarity structure, (c) yields predictions that have a high level of fit with the observed data. Possible implications from the use of the model for future research are briefly discussed.