Two commonplace assumptions about encoding are that sentences are encoded and recognized on the basis of their semantic features primarily and that information regarding form features such as typography is typically ignored or discarded. These assumptions were tested m the present experiment where, within a signal-detection paradigm, S sorted sentences according to whether he had seen them before or not (old vs new) and, if they were old, whether their reappearance was in the same typography as on the first occurrence or a different one. Of the two typographies, one was familiar and the other unfamiliar. Results show that a considerable amount of information regarding surface features is stored for many minutes and that ease of initial encoding is inversely related to likelihood of subsequent recognition: sentences in the unfamiliar typography were remembered better. The results are probably not due to time spent encoding; control tests suggest that time spent encoding a difficult typography does not by itself increase recognition of the semantic content embodied in the typography. Other control tests show that pictorial features or images of the sentences play no significant role in their subsequent recognition. One interpretation of the results is that the analytic activities or cognitive operations that characterize initial acquisition play a significant role in subsequent recognition.