Background: Results from several studies indicated that a symptom model other than the DSM triad might better describe symptom domains of autism. The present study focused on a) investigating the stability of a new symptom model for autism by cross-validating it in an independent sample and b) examining the invariance of the model regarding three covariates: symptom severity, intelligence, and age. Method: The validity of the symptom model was examined in an independent sample of N = 263 children and adolescents with autism spectrum disorders, and model invariance was studied in a larger sample of N = 356 children and adolescents with autism spectrum disorders. The fit of the symptom model to the sample data was compared to that of alternative models (including the DSM triad), and the invariance of the new model was investigated for each covariate by multiple-group comparisons. Results: The fit of the new symptom model was better than that of two alternative models. It could not be compared to that of the DSM triad, because the latter encountered empirical identification problems. There were no significant or substantive differences between the estimated model in each of the dichotomised groups for any of the three covariates, which indicated factorial invariance of both structural form and factor loadings. Conclusions: The symptom model appeared to be relatively stable: It could be cross-validated in the independent sample and factorial invariance was shown between the dichotomised groups for each covariate. Further model validation with instruments other than the Autism Diagnostic Interview-Revised (ADI-R) is recommended.