The reproducibility of an interactive image registration technique used as part of the radiotherapy treatment planning process was investigated for 3D CT and MR pediatric head images. Over a nine month period, 85 CT/MR image registrations, required for treatment planning, were repeated, 52 by the same operator and 33 by a different operator. All were performing image registrations for normal clinical care and the first registration was used clinically. Inter- and intra-operator reproducibility of the translation and rotation were calculated separately. The standard deviation of the average total translation and rotation was 0.39 mm and 1.7 degrees, and 0.58 mm and 2.8 degrees, respectively. The maximum difference between registrations was 1.1 mm and 4.1 degrees when repeated by the same operator, and 1.4 mm and 5.8 degrees when repeated by another operator. The variation for the lowest resolution parameters, out of plane translation and rotations, was 2 to 3 times larger than for in-plane movements. A registration took between 5 minutes and over half an hour for difficult cases, with a mean of 14.3 minutes. One to two millimeter reproducibility was not achieved and interactive registration was relatively time consuming. There is a clear image resolution effect on registration reproducibility, suggesting that reducing slice thickness could considerably improve registration reproducibility.