Autism spectrum disorders (ASD) are a highly heterogeneous group of neurodevelopmental disorders that are more commonly diagnosed in boys than in girls. The reasons for gender differences in ASD are unknown and no definitive current evidence can explain male predominance. Therefore, in search for laboratory biomarkers responsible for ASD, a comprehensive metabolomics study was performed by metabolic profiling of urine samples in 51 ASD subjects and 51 age- and sex-matched children with typical development. Orthogonal partial least-squares discriminant analysis (OPLS-DA) models with poor quality failed to perform the analysis based on gender in the ASD and control groups. OPLS-DA models based on single-sex samples, especially in female subjects, had better clustering between the ASD and control groups with an increase in the R and Q values compared with those in the whole group. Significantly increased levels of adenine, 2-Methylguanosine, creatinine, and 7alpha-hydroxytestololactone and a decrease in creatine were observed in the female ASD subjects. In particular, 7alpha-hydroxytestololactone, which has a structure similar to that of testolactone, was positively correlated with adenine (Pearson correlation coefficient, r = 0.738, p < 0.01), creatinine (r = 0.826, p < 0.01), and 2-Methylguanosine (r = 0.757, p < 0.01) and negatively correlated with creatine (r=-0.413, p < 0.05). A receiver operating characteristic curve analysis using the creatinine:creatine ratio yielded an area under the curve of 0.913 (95% CI: 0.806-1). These metabolites may be sex-related or sex-sensitive to an extent and can be valuable for identification of the molecular pathways involved in the gender bias in manifestation of ASD. The creatinine:creatine ratio has a potential to be a good predictor of ASD in the female subjects.