Program understanding tools that are able to process large legacy systems are needed to support software engineers with maintenance tasks. One type of tool, a plan-based program understander, recognizes concepts in the code by comparing the target code against a library of programming plans. For a large plan library, an exhaustive search for a matching plan during concept recognition is not effective. This drawback limits the practical scalability of these types of tools. We propose a signature-based approach to prune the plan library. A signature in this work is a set of high level attributes derived from code features suggested by studies on human program comprehension. A SIGMA prototype is implemented to embody the proposed approach. Program plan retrieval with SIGMA show promising results in removing irrelevant plans from consideration prior to attempting the later computational concept recognition process. It appears that the signature approach to plan retrieval may be able to facilitate the scaling up of plan-based understanders so that they can better support maintenance tasks on large legacy systems.