A method is developed for automatic analysis of the reliability of an automatic registration of perfusion cardiovascular MR images. A parameter, for example, a similarity measure between the successive images, is calculated first in order to quantify the success of the registration process between these images for the data set. A criterion is then introduced, for example a threshold is imposed on the similarity measure. The successively registered images that have a calculated similarity measure that exceeds the defined threshold are automatically accepted for further analysis.