The authors introduce a multifactor dimensionality reduction (MDR) method and software for detecting gene–gene and gene–environment interactions in complex diseases. MDR is designed to address the challenges of high-dimensional genetic data and small sample sizes by collapsing high-dimensional genetic data into a single dimension, allowing for the detection of interactions. The software integrates MDR with cross-validation to estimate classification and prediction errors of multifactor models. It can analyze interactions among 2–15 genetic and/or environmental factors, handling datasets with up to 500 total variables and 4000 study subjects. The method is model-free and nonparametric, making it suitable for various genetic models. The authors demonstrate the effectiveness of MDR through simulations and real-world studies, showing its ability to identify high-order interactions in the absence of significant main effects. The software is available for Linux and Sun platforms, and the input format supports both text and pre-makedep formats. The article also discusses limitations and future directions, including the use of machine learning strategies and the extension of MDR to large, complex pedigrees.The authors introduce a multifactor dimensionality reduction (MDR) method and software for detecting gene–gene and gene–environment interactions in complex diseases. MDR is designed to address the challenges of high-dimensional genetic data and small sample sizes by collapsing high-dimensional genetic data into a single dimension, allowing for the detection of interactions. The software integrates MDR with cross-validation to estimate classification and prediction errors of multifactor models. It can analyze interactions among 2–15 genetic and/or environmental factors, handling datasets with up to 500 total variables and 4000 study subjects. The method is model-free and nonparametric, making it suitable for various genetic models. The authors demonstrate the effectiveness of MDR through simulations and real-world studies, showing its ability to identify high-order interactions in the absence of significant main effects. The software is available for Linux and Sun platforms, and the input format supports both text and pre-makedep formats. The article also discusses limitations and future directions, including the use of machine learning strategies and the extension of MDR to large, complex pedigrees.