Two methods for 3D realignment of activation brain single-photon emission tomographic (SPET) studies are analyzed. The first is based on principal axes transformation (PAT). The second uses the results of the first method as initial values to start a least-squares iterative process (LS) to search for the maximum value of the correlation function. Both methods were tested with simulated and real studies. The results of the PAT method showed a maximum translation error of 0.3 +/- 0.1 pixels and a rotational error of 1.2 +/- 0.7 degrees in a total of 100 runs. For the LS method these errors were 0.2 +/- 0.1 and 0.6 +/- 0.3. The realignment for 34 real studies was assessed by three expert observers. The alignment was found to be satisfactory in all cases for the LS method, and in 18 cases (53%) for the PAT method. From the results we conclude that a combination of both methods allows the accurate realignment of SPET neuroactivation studies.