Abstract: The goal of the Multiple-Choice Reading Comprehension (MCRC) task is to require a model to select the correct answer from a set of candidate options based on the given paragraph and question. Related work shows that although large language models (LLM) have achieved state-of-the-art performance on this task, the failure to activate option-related thought prevents them from correctly perceiving the deep connections between the passage, question, and options. To this end, we propose a LLM enhancement framework for MCRC based on Question-Solving Thought (QST). Specifically, two types of mechanism are considered in our work: step-by-step and option-by-option. Two methods are further proposed to better utilize these QSTs: Self Activate (SA) and Distill from Teacher (DT), which further boost the question-solving of the model. Experimental results demonstrate that our proposed approaches increase average accuracy and average accuracy-hard by 22.61% and 28.95% respectively, compared to existing baselines across six multiple-choice reading comprehension datasets.