We explore familiarity-based recognition using a paradigm devised by Parkin et al. (2001). The task consists of the creation of two lists of words written with one of two different subsets of letters of the alphabet. We manipulated study time (50, 100, 200, 500 ms per word) of words with different letter probabilistic structure to those originally used by Parkin et al. Letter-based familiarity responding was robust and present even at rates producing otherwise chance performance. A second experi- ment and structural equation modelling led us to interpret the results from the point of view of a theory that takes into account the processing of similarities and differences (Hunt & MacDaniel, (1993). Finally, our data indicate that the experimental procedure devised by Parkin et al. is an excellent tool with which to study familiarity, once the structure of probabilities of individual letters is considered as the key factor in inducing the effect.