This study compared the effectiveness of two reading treatment programmes for two contrasting developmental dyslexics. W.B. demonstrated "pure" phonological dyslexia (deficient nonword reading but normal irregular-word reading) and N.S. "pure" surface dyslexia (the converse pattern). Both participants completed: (a) a phonological programme, which targeted the sublexical reading procedure through repeated exposure to word "families" with the same grapheme-phoneme correspondence (GPC; e.g., frail, raid, bait); and (b) a whole-word programme, which targeted the lexical reading procedure through tasks that emphasize whole-word visual analysis (e.g., speeded identification of visually degraded words). Both participants improved after training on the targeted words and/or GPCs. However, W.B. demonstrated reliable generalization only following the phonological programme and only in his reading of nonwords. In contrast, N.S. showed generalization across all types of word materials following both programmes. Although the whole-word programme (in particular the degraded-images technique) resulted in numerically greater improvement for N.S., this difference was not significant. Practical and theoretical implications of these results are discussed.