This paper seeks to present an alternative to existing cost driver research by applying artificial neural networks and traditional regression analysis methods to identify the relationship between manufacturing overhead costs and cost drivers. To this end, the study evaluates whether predictive performance improves when nonlinearity in functional relationships is permitted by comparing the prediction performance of manufacturing overhead costs using simple linear regression equations versus artificial neural networks that allow for nonlinearity, and analyzes whether the inclusion of complexity-related cost drivers advocated by Activity-Based Costing (ABC) improves predictive power for manufacturing overhead costs. Existing studies related to cost drivers have mostly used regression models with manufacturing overhead as the dependent variable and cost drivers as independent variables, determining cost driver status based on the statistical significance of cost driver variables. This study is distinctive in that it determines cost driver status based on the predictive performance for manufacturing overhead costs as the dependent variable. The research results revealed that volume-related variables can serve as useful criteria for predicting manufacturing overhead in both regression models and artificial neural networks. This implies that there is no significant problem with using the volume-related criteria traditionally employed in manufacturing overhead budgeting. Moreover, the results indicate that the predictive power of linear regression equations commonly used in flexible budgeting is not inferior to that of artificial neural networks, which are capable of capturing nonlinear functional relationships.