Objective Searching for objective biomarkers of autism spectrum disorders (ASD) can provide assisted diagnosis for doctors.Resting-state functional connectivity (RSFC) reflects temporal correlation of the neuron activity patterns between different brain regions.Researchers often explore biomarkers for identifying ASD from RSFCs.However,most approaches cannot select discriminative RSFCs effectively.In this paper,we apply least absolute shrinkage and selection operator (Lasso) to select most discriminative RSFCs between ASD children and typically developing (TD) children.Methods Firstly,RSFCs were extracted as features by using Pearson correlation analysis,and were thresholded to retain RSFCs with larger positive correlation values.Then,Lasso method was adopted to select discriminative features from RSFCs.Finally,we used support vector machine to identify ASD children from TD children and mainly took classification accuracy as index to evaluate classification performance.Results The method based on Lasso achieved an accuracy of 81.52%,and the discriminative RSFCs had significant difference between ASD and TD children.Conclusions This method improves the classification accuracy of ASD,and the biomarkers have the potential to be applied in clinical diagnosis.