A method is presented for prediction of the systemic drug concentration profile from in vitro release/dissolution data for a drug formulation. The method is demonstrated using four different tablet formulations containing 200 mg carbamazepine (CZM), each administered in a four way cross‐over manner to 20 human subjects, with 15 blood samples drawn to determine the resulting concentration profile. Amount versus time dissolution data were obtained by a 75 rpm paddle method for each formulation. Differentiation, with respect to time, of a monotonic quadratic spline fitted to the dissolution data provided the dissolution rate curve. The dissolution curve was through time and magnitude scaling mapped into a drug concentration curve via a convolution by a single exponential, and the estimated unit impulse response function. The method was tested by cross‐validation, where the in vivo concentration profiles for each formulation were predicted based on correlation parameters determined from in vivo–in vitro data from the remaining three formulations. The mean prediction error (MPE), defined as the mean value of 100% x(observed−predicted)/observed was calculated for all 240 cross‐validation predictions. The mean values of MPE were in the range of 10–36% (average 22%) with standard deviations (S.D.s) in the range of 9–33% (average 13%), indicating a good prediction performance of the proposed in vivo–in vitro correlation (IVIVC) method. Copyright © 2000 John Wiley & Sons, Ltd.