the scheduler is able to predict the average service time of the these investigations and decide the subclass. On the other hand, some investigations have been grouped together (elbow, shoulder and foot, hip, neck). These are those investigations that were performed rarely (small sample size) and shown similarity in service times and hence grouped together. After the broad classification as mentioned above, the CART analysis was performed in order to divide the patients into a num ber of classes using the data on source of arrival of patient, the broad category of investigation and the service times for the sample of patients. CART analysis was performed using the "classregtree' function available in the statistical toolbox in MATLAB that used the CARTalgorithm as described in [31] The full tree obtained by the algorithm provided 21 classes (with the condition that minimum number of data points in a child node is 10). In order to have a manageable number of classes of patients having significant difference in their characteristics, pruning was performed to finally have six patient classes (see Table 3). The last column of Table 3 contains the coefficient of variation (CV) of the patient classes de- fined as the ratio of standard deviation to mean of service time which is used to measure the service time variability. The service time distributions for these classes of patients are given in Table 4.