Background. - The prevalence of poor reading skills is particularly high among children from low socioeconomic backgrounds, but no longitudinal studies have been conducted so far in France to determine whether poor reading in a socioeconomically challenged population is persistent and warrants preventive action. Design. - One hundred and fifty-four children were divided into three groups according to their reading skills: poor, intermediate and typical readers. They were followed over a period of 2 years. Reading levels, spelling and comprehension were assessed by standardized measurement scales in order to determine reading outcome and predictive variables. Results. - The reading skills in each group progressed at similar rates, but the differences between the three groups remained relatively constant over the 2 years. The gap between good and poor readers actually increased for the poorest readers. Spelling scores followed a similar pattern and remained weak. Comprehension scores followed a different pattern. Most of the initially poor readers with low comprehension scores almost caught up and reached the level of the typical readers. The best predictive variables of reading and spelling outcome were phonological awareness, rapid naming and attention deficit. The strongest predictive variables for comprehension were IQ, lexical level and attention. Conclusions. - Our results confirm the relative stability of reading measurement across time in poor readers from low socioeconomic backgrounds. Their behavior are similar to the classic dyslexic population. The predictive variables are different depending on whether reading or spelling or comprehension is considered. These results provide a clear agenda for preventive literacy action in children with low socioeconomic levels (SES): phonological decoding and oral language skills in early grades, and screening and treatment of attention disorders. (C) 2010 Elsevier Masson SAS. All rights reserved.