Call history data is sampled at fixed intervals during run-time, each sample representing only a limited portion of the stack. These data samples are subsequently automatically analyzed by merging overlapping sampled call history sequences to build larger call graphs, according to some pre-specified merge criterion. Preferably, the call history graphs are annotated with counts of the execution frequency (number of times a particular procedure was executing when the sample was collected) and the stack frequency (number of times the procedure appeared in the sampled stack portion) associated with each respective called procedure. Preferably, a graphical user interface presents the user with a graphical representation of the call graph(s) and annotations.