Selecting appropriate reading materials for L2 (second language/foreign language) learners is crucial for improving their proficiency in the target language. However, the limitation of effective Chinese text readability classifiers poses a significant hurdle for students and educators in accurately gauging the precise difficulty level of texts in international Chinese education. This research conducted the readability grading of Chinese as a Second Language (CSL) texts by developing a BERT-Based CSL Readability Classifier (BCRC), which utilizes the BERT architecture specifically trained on CSL texts and incorporates multidimensional linguistic features including lexical richness, syntactic complexity and syntax patterns. The model was evaluated using a dataset of CSL texts, and the results indicate that the BCRC model performs effectively in predicting the readability levels of CSL texts. It achieves high mean accuracy of 92.9% across different readability levels, which outperforms baseline classifiers in terms of classification performance, highlighting the enhancement capabilities of multidimensional linguistic features in CSL readability classification models. This study contributes to the field of CSL education by providing a robust readability classifier as a valuable tool for educators, curriculum designers, and developers of CSL learning materials to ensure appropriate text selection based on learners' proficiency levels.