The evaluation of simulation models is a fundamental step in establishing their credibility. Conventional approaches often rely on qualitative expert judgment, a process that is often inefficient and labor-intensive. Additionally, simulation outputs are often limited due to high computational costs or intrinsic model properties. To address these challenges, we propose a Quantitative Learning Method (QLM) that employs a dual-layer architecture. This method is designed to integrate historical assessment scores in small-sample contexts, learning the mapping from raw simulation outputs to evaluation metrics. Validated on two classic simulation models against seven baseline algorithms, QLM demonstrates robust generalization and stable performance, mitigating the overfitting prevalent in machine learning approaches under data constraints. Furthermore, its inherent transparency the mapping between simulation outputs and credibility scores offers considerable interpretability and practical value.