Defining a quantitative and reliable relationship between in vitro drug release and in vivo absorption is highly desired for rational development, optimization, and evaluation of controlled-release dosage forms and manufacturing process. During the development of a once-daily extended-release (ER) tablet of divalproex sodium, a predictive in vitro drug release method was designed and statistically evaluated using three formulations with varying release rates. In order to establish an internally and externally validated Level A IVIVC, a total of five different ER formulations of divalproex sodium were used to evaluate a linear IVIVC model based on the in vitro test method. For internal validation, a single-dose four-way crossover study (N = 16) was performed using fast-, medium-, and slow-releasing ER formulations and a 12-h IV infusion of valproic acid as reference. To validate the IVIVC externally, a second three-way crossover study (N = 36) was performed using slightly-fast-, medium-, and slightly-slow-releasing ER formulations. The in vivo absorption-time profile was inferred by deconvolution of the observed plasma concentration-time profiles against the unit disposition function (UDF). A linear IVIVC model was established in which the in vivo absorption was expressed as a function of in vitro drug release. Plasma profiles of ER formulations were estimated via convolution of in vitro release profiles with the UDF. Successful internal and external validations of the model were demonstrated by individual and average absolute percent prediction errors of </=9% for both C(max) and AUC(infinity). In conclusion, a Level A IVIVC describing the entire time-course of plasma concentrations was developed and validated, both internally and externally, for ER formulations of divalproex sodium.