Research in human genetics has recently focused on investigating complex diseases that may be controlled by both genetic and environmental factors, possibly interacting together. A recursive-partitioning (RP) algorithm has been developed for identifying subgroups of affected relative pairs demonstrating increased evidence for linkage, where subgroups are defined by pair-level or family-level covariates. After growing a full tree, optimal tree structures are identified through a tree pruning procedure and a final tree structure selection procedure by use of bootstrap algorithms. Simulation studies showed that power to detect linkage can increase in the presence of gene-environment interactions, depending on the magnitude of the interaction, and the RP linkage model correctly identifies covariates associated with disease genes in a large proportion of simulations. The RP linkage model was applied to a dataset of families with bipolar disorder where linkage regions on chromosome 18 have been previously identified. Using the RP model, several suggestive regions were found on chromosome 18, and associated environmental factors were detected. The RP model has the potential to identify previously unknown gene-environment interactions, and the practical utility and potential of this new methodology have been demonstrated.In addition, for multiple dependent affected relative pairs, a GEE linkage approach was proposed, and was compared to an analysis using the log-likelihood ratio (LLR) statistics from a likelihood ratio (LR) linkage model (Olson 1999) with significance thresholds set empirically. Properties of quasi-likelihood ratio statistics from the GEE linkage model were studied by simulations. In all the simulation studies, based on an empirical approach, the LLR statistics from the LR linkage model with an independence assumption had better power to detect linkage, and were recommended as goodness of split statistics for the RP linkage model.