We present a novel approach to compute a collision-free path for part disassembly simulation and virtual prototyping of part removal. Our algorithm is based on sample-based motion planning that connects collision-free samples in the configuration space using local planning. In order to effectively handle the tight-fitting scenarios, we describe techniques to generate samples in narrow passages and efficient local planning algorithms to connect them with collision-free paths. Our approach is general, makes no assumption about model connectivity or object topology, and can handle polygon soup models that frequently arise in CAD applications. We highlight the performance on many challenging benchmarks including the Alpha puzzle, maintainability of the windscreen wiper motion, and disassembly of a seat from the interior of a car body.