Background: Writing is a complex activity involving various cognitive processes in the planning, the transcription and the revision of written texts. The present study focused on the revision of written texts within a developmental approach. Aims: The study aimed to examine whether children and adults use different procedures to detect and revise erroneous grammatical agreements. It was predicted that children would use a slow algorithmic procedure while adults would use a fast automatized procedure. Sample: One hundred and twenty participants from 5th grade to undergraduate levels (24 per level) participated in the study. Method: The participants were asked to decide as quickly as possible whether a visually presented sentence had any agreement error. The French experimental sentences were of the type 'The N1 of the N2 + Verb', in which N2 was either a plausible subject of the following verb (e.g., The guard of the prisoners watches) or an implausible subject (e.g., The guard of the safes watches). Correctness and latency of the responses were recorded. Results: The main results showed that only the younger participants were affected by the subject-role plausibility of N2, and that there was no difference in response latency between their correct and incorrect responses. These observations support the hypothesis that the younger participants systematically apply a time-consuming algorithmic procedure to verify the agreement; since one step of this procedure consists in searching for the subject of the verb, these participants were frequently misled by the subject-role plausibility of N2. On the contrary, the older participants were not affected by the plausibility of N2, but were frequently misled by erroneous agreements between N2 and the verb. These observations support the view that these older participants use a fast decision strategy based on the co-occurrence of formal indices. Their correct answers, however, were slower than their incorrect ones; this suggests that they also sometimes use a time-consuming controlled procedure. Conclusion: The study shows that along with the acquisition of writing expertise, the revising activity itself is progressively facilitated and gradually automatized by substituting a fast direct decision strategy for a slow and laborious use of revision rules.