Autism spectrum disorder (ASD) is a neurodevelopmental disorder closely related to potential dysfunction of the brain. Although multiple functional connectivity (FC) features such as low-order functional connectivity (LOFC) and high-order functional connectivity (HOFC) provide complementary knowledge to each other, it is still challenging to find interpretable LOFC and HOFC features for multi-center ASD diagnosis. To this end, we develop a novel interpretable feature learning method based on multi-output TSK fuzzy system (MO-TSK-FS) for multi-center ASD diagnosis. Specifically, both the LOFC and HOFC features are first mapped to a high-dimensional space using the premise part of MO-TSK-FS, which shares the common knowledge across multiple centers. Then, the mapped features are transformed to a low-dimensional feature space using a transformation matrix. A novel unsupervised learning problem is formulated to find the optimal transformation matrix. Finally, a multi-modality support vector classifier (M2SVC) is constructed for classification. The experimental results show that the proposed interpretable feature learning method for multi-center ASD classification can effectively extract important features from the original LOFC and HOFC features, resulting in an efficient M2SVC for multi-center ASD classification.