In order to display the surfaces of internal structures within a solid body from non-intrusively acquired data sets, it is useful to segment the data sets into the internal structures of interest before searching for the surfaces of such structures. To accomplish this, a data segmentation system uses a plurality of sample data points to construct a statistical probability distribution for a plurality of internal structures. Using these probability distributions, each data point is labeled with the most likely structure identification. Searching the thus-segmented data points for surfaces is considerably faster than is possible with the entire data set and produces surface renditions with fewer anomalies and errors. If the solid body is a human head, and nuclear magnetic resonance (NMR) is used to obtain two data sets corresponding to the two NMR echoes, then the probability distribution is bivariate and the two echoes can be plotted against each other to assist in identifying tissue clusters.