The present study examined training effects in dyslexic children on reading fluency and the amplitude of N170, a negative brain-potential component elicited by letter and symbol strings. A group of 18 children with dyslexia in 3rd grade (9.05 ± 0.46 years old) was tested before and after following a letter-speech sound mapping training. A group of 20 third-grade typical readers (8.78 ± 0.35 years old) performed a single time on the same brain potential task. The training was differentially effective in speeding up reading fluency in the dyslexic children. In some children, training had a beneficial effect on reading fluency (‘improvers’) while a training effect was absent in others (‘non-improvers’). Improvers at pre-training showed larger N170 amplitude to words compared to non-improvers. N170 amplitude decreased following training in improvers but not in non-improvers. But the N170 amplitude pattern in improvers continued to differ from the N170 amplitude pattern across hemispheres seen in typical readers. Finally, we observed a positive relation between the decrease in N170 amplitude and gains in reading fluency. Collectively, the results that emerged from the present study indicate the sensitivity of N170 amplitude to reading fluency and its potential as a predictor of reading fluency acquisition.