Autism spectrum disorder (ASD) is a highly heterogeneous neurodevelopmental disorder. The estimation of ASD severity is very important in clinical practice due to providing a more elaborate diagnosis. Although several studies have revealed some resting-state functional connectivities (RSFCs) that are related to the ASD severity, they have all been based on small-sample data and local RSFCs. The aim of the present study is to adopt multivariate pattern analysis to investigate a subset of connectivities among whole-brain RSFCs that are more contributive to ASD severity estimation based on large-sample data. Regression estimation shows a Pearson correlation value of 0.5 between the estimated and observed severity, with a mean absolute error of 1.41. The results provide obvious evidence that some RSFCs undergo notable alterations with the severity of ASD. More importantly, these selected RSFCs have an abnormality in the connection modes of the inter-network and intra-network connections. In addition, these selected abnormal RSFCs are mainly associated with the sensorimotor network, the default mode network, and inter-hemispheric connectivities, while exhibiting significant left hemisphere lateralization. Overall, this study indicates that some RSFCs suffer from abnormal alterations in patients with ASD, providing additional evidence of large-scale functional network alterations in ASD.