One of the most important solutions for dimensionality reduction in data preprocessing, and improving classification performance is gene selection in microarray data since they usually have several thousand genes with very few samples. Because of these characteristics, the complexity of classification models increases and their efficiency decreases. The gene selection problem inherently pursues two goals: reducing the number of genes and increasing the classification efficiency. Therefore, this paper presents a novel hybrid filter-wrapper solution based on the Fisher-score method and Multi-Objective Forest Optimization Algorithm (MOFOA). In the proposed method, as a preprocessing step, the Fisher-score method selects 500 discriminative genes by removing redundant/irrelevant genes. Then, MOFOA searches to find the subset of optimal genes using concepts such as repository, crowding-distance, and binary tournament selection. Moreover, the proposed method solves the gene selection problem and, at the same time, optimizes the kernel parameters in the SVM classification model. Six microarray datasets were used to evaluate the performance of the proposed method. Afterward, a comparison was made between its results and those of the four multi-objective hybrid methods presented in the literature in terms of classification performance, the number of selected genes, running time, and hypervolume criteria. According to the results, in addition to selecting fewer genes, the proposed solution has achieved greater classification accuracy in most cases and has been able to obtain a performance similar to or better than that of other multi-objective gene selection approaches.