Data analysis in single molecule studies often involves estimation of parameters and the detection of abrupt changes in measured signals. For single molecule studies, tools for automated analysis that are crucial for rapid progress, need to be effective under large noise magnitudes, and often must assume little or no prior knowledge of parameters being studied. This article examines an iterated, dynamic programming based step detection algorithm (SDA). It is established that given a prior estimate, an iteration of the SDA necessarily improves the estimate. The analysis provides an explanation and a confirmation of the effectiveness of the learning and estimation capabilities of the algorithm observed empirically. Further, an alternative application of the SDA is demonstrated, wherein the parameters of a worm-like chain (WLC) model are estimated, for the automated analysis of data from single molecule protein pulling experiments.