This discussion is based on the results of an earlier experiment in which four groups of deaf subjects, ranging in age of first exposure to signing from birth to over eighteen, were given lists of sentences in American Sign Language to shadow and recall immediately after presentation. It was found that in terms of overall accuracy, early learners outperformed late learners, that sentences were recalled and shadowed more accurately than scrambled sentences, and that shadowing was more accurate than immediate recall. This paper concentrates on an analysis of the errors in that experiment, since patterns of errors can give more clues as to how people process language. Three main categories of errors were classified--deletion, addition, and substitution. Addition and substitution errors were further classified into whether they were random, repetitions, semantic, phonological, or simultaneously semantic and phonological. Groups I (native) and II (5-7) pattern very similarly in both the number of errors and in the proportion of errors in the various categories. Those in group IV (18 plus) seem lost while those in group III (13-15) pattern similarly. Sentencehood has an effect on the performance of only the first two groups. The specific task leads to differential error patterns across the same groupings. (Author/JK)