Background: Event-Related Potentials (ERPs) permit to study neuronal specialization during reading acquisition. The N170 wave was previously shown to be a surrogate of the fine tuning of reading in adults and adolescents as well. Aim: We analyzed and described the variations of the N170 wave as a function of French words with visual or phonetical similarities in 12 to 14 yearold dyslexic patients. We tested the validity and modulation of this effect by comparing different populations of normal and dyslexic patients of various severity. Methods: ERPs were recorded in seventeen dyslexic children with the same method as in normative populations in lexical decision. Stimuli consisted of frequent words chosen on the basis of near or far visemes and morphemes. Dyslexic children were compared to two control (i.e. normal readers) groups, one group of the same age (N=15) and one group of adults (N=17). N170 and P100 waves were analyzed, as well as interactions between both (i.e. P1N1) searched. Psychometric and language tests were also performed. Results were analyzed by ANOVA. Results: The results of sixteen patients are presented. All sixteen showed significant differences on all psycholinguistic items when compared to the two control groups. All groups (patients and controls) significantly differed from each other for all tests (F'(4;42)=119, 2; p<0.001). However, the heterogeneity of the ERP patterns in the dyslexic group rendered group averaging irrelevant. The N170 wave sometimes overlapped with the P100 wavelength, but was also found to be negative or absent in some patients. In the course of development, N170 variations seem dependent on characteristics of the P100. No correlation was found between the variations of the N170 and clinical measures. Analysis of the P1-N1 temporal course showed a tendency to correlate with reading speed for the entire study population, yet not reaching statistical significance. Interpretation: N170 variations during development and dyslexic pathologies are associated with P100 variations. The P1-N1 time course could reflect silent reading speed. The P1-N1 temporal course was linked with clinical measures in all three groups, which could reflect neurodevelopmentalyrelated variations of the heterogeneity of the N170 as well as a developmental pathology. Verbal stimuli permit us to test the N170 physiological heterogeneity during development but variations in response to easy tasks show low sensitivity of N170 as a marker of dyslexia.