Background: The conclusion from a vast literature on literacy acquisition is that letter knowledge is one of the best predictors of literacy development. The question of the best way to teach children letter sounds has not, as yet, been answered satisfactorily. Aims: The aim of this study was the evaluation of a computer training program using integrated-picture mnemonics combined with a fading procedure to teach children letter sounds. Sample: Thirty-nine kindergartners attending mainstream primary education participated in this study. Method: A within-subject design was used. Each kindergartner learned letters under three conditions: (a) a fading condition in which letters are taught using a picture-supported first-sound mnemonics procedure in combination with a fading procedure; (b) an embedded condition in which letters are taught using the picture-supported first-sound-mnemonics procedure only and (c) a without-picture condition in which letters are taught using a first-sound procedure without-picture support. Dependent measures included a productive and receptive letter-sound test, and a first-sound isolation task. Results: Productive letter-sound knowledge in the fading condition was better than in the other two conditions. In addition, kindergartners with good and those with poor first-sound isolation ability performed equally well in the fading condition. However, in the embedded and in the without-picture conditions, the kindergartners with good first-sound isolation ability outperformed those with poor isolation ability. Conclusion: These findings indicate that an integrated-picture mnemonics procedure combined with a fading procedure is effective in teaching kindergartners letter sounds and that the success of such a procedure does not depend on their initial first-sound isolation ability.