Textbooks are essential resources for classroom and offline reading, while the quality of learning materials guides the entire learning process. One of the most important factors to be considered is their readability and comprehensibility. Therefore, the correct pairing of textbook complexity and student grade level is paramount. This article analyzes automated classification methods for Russian-language textbooks on two dimensions, namely the topic of the text and its complexity reflected by its corresponding school grade level. The studied corpus is a collection of 154 textbooks from the Russian Federation from the second to the eleventh grade levels. Our analysis considers machine learning techniques on the textual complexity indices provided by the open-source multi-language framework ReaderBench and BERT-based models for the classification tasks. Additionally, we explore using the most predictive ReaderBench features in conjunction with contextualized embeddings from BERT. Our results argue that incorporating textual complexity indices improves the classification performance of BERT-based models on our dataset split using 2 strategies. More specifically, the F1 score for topic classification improved to 92.63%, while the F1 score for school grade-level classification improved to 54.06% for the Greedy approach in which multiple adjacent paragraphs are considered a single text unit up until reaching the maximum length of 512 tokens imposed by the language model.