A traditional technique for structure-based drug design (SBDD) is mapping of protein surfaces with probe molecules to identify "hot spots" where key functional groups can best complement the receptor. Common methods, such as minimization of probes or calculation of grids, use a fixed protein structure in the gas phase, ignoring both protein flexibility and proper competition between the probes and water. As a result, the potential surface is quite rugged, and many spurious local minima are identified. In this work, we compared rigid and fully flexible proteins in mixed-solvent molecular dynamics, which allows for flexibility and full solvent effects. We were surprised to find that the large number of local minima are still found when a protein's conformational sampling is restricted; the dynamic averaging of probes and competition with water do not smooth the potential surface as one might expect. Only when a protein is allowed to be fully flexible in the simulation are the proper minima located and the spurious ones eliminated. Our results indicate that inclusion of full protein flexibility is critical to accurate hot-spot mapping for SBDD.