Autism spectrum disorder (ASD) is a neurodevelopmental disorder that begins early in life and continues lifelong with strong personal and societal implications. It affects about 1%-2% of the children population in the world. The absence of auxiliary methods that can complement the clinical evaluation of ASD increases the probability of false identification of the disorder, especially in the case of very young children. In this study, analytical models for auxiliary diagnosis of ASD in children and adolescents, based on the analysis of patients' blood serum ATR-FTIR (Attenuated Total Reflectance-Fourier Transform Infrared) spectra, were developed. The models use chemometrics (either Principal Component Analysis (PCA) or Partial Least Squares Discriminant Analysis (PLS-DA)) methods, with the infrared spectra being the X-predictor variables. The two developed models exhibit excellent classification performance for samples of ASD individuals vs. healthy controls. Interestingly, the simplest, unsupervised PCA-based model results to have a global performance identical to the more demanding, supervised (PLS-DA)-based model. The developed PCA-based model thus appears as the more economical alternative one for use in the clinical environment. Hierarchical clustering analysis performed on the full set of samples was also successful in discriminating the two groups.