It might be stuck in a local optima, some things to check: A) Read the section "score trap" in the docs, to verify you don't have a score trap B) The default move types (<changeMoveSelector> and <swapMoveSelector>) aren't strong enough to escape local optima. B1) Try adding a <pillarSwapMoveSelector>. I am adding a pillarChangeMove for 6.1.0.Beta4, and probably also support for selecting subpillars. B2) Take a look at that suboptimal best solution and figure out what course-grained move it needs to do to escape the local optima. Then either creates a custom Move (see docs "MoveListFactory") B3) or compose the build-in ones with <cartesianProductMoveSelector> and mimicSelection. C) For very small datasets, tabu search's entity size can be too large (especially with subchain move selectors and pillar move selectors). Tabu ratio (see docs) is a way around that. D) If your dataset is really small, you can always just switch to brute force or the much faster branchAndBound (as of 6.1.0.Beta3). Note: neither scale to medium/big datasets. HTH On 02-05-14 17:06, nkgxgongxi wrote: