The R aregImpute Function for Flexible Multiple Imputation Frank E Harrell Jr Department of Biostatistics, Vanderbilt University School of Medicine Flexible modern regression methods that relax the traditional regression assumptions are being used with increasing frequency. There is a parallel need for flexibility in multiple imputation models. There is also value in ensuring reasonableness of distributions of imputed values, leading to increased usage of predictive mean matching (PMM) as the basis for imputation. The aregImpute function in the R and S-Plus Hmisc package uses regression and PMM, incorporating parametric least squares generalized additive models so as to not assume linearity at any point. Canonical variates can be used to automatically transform a continuous or polytomous variable when predicting it for the purpose of imputing its missing values. The approximate Bayesian bootstrap is central to the approach. For PMM, weighted multinomial sampling is used to avoid overuse of a donor observation, ensuring a smooth distribution of imputed continuous variables. Idea for JSS for Recai Yucel