Mathematical problem-solving is a core content of primary school education. Therefore, it is necessary to provide all students with a meaningful way of acquisition of problem-solving skills. The objective of this research was to verify the effectiveness of a problem-solving routine based on the Pólya's model and the use of cognitive strategies combined with systematic and explicit instruction with continuous feedback, to improve additive problem-solving skills of students with autism. Twenty-eight students with autism participated in a quasi-experimental exploratory study. The empirical study consisted of four phases: pretest, intervention, posttest, and maintenance. The results show a significative improvement of the students' performance in the experimental group after the intervention, and most of them maintained what they have learned six weeks after the intervention. Between-group effect size during the post-test supports the effectiveness of the intervention.