The watershed transform is one of best known and widely used methods for image segmentation in mathematical morphology. Since the definition, deriving from geology and nature observation is quite intuitive and straightforward to implement; many fast and powerful algorithms for watershed transform have already been presented. However, there still occur problems when one wishes to achieve a precise solution on blurred or noised image. The same range of problems is faced when a plateau occurs in the image. In this paper several methods for plateau reduction are discussed and some novel ideas proposed. All algorithms are performed on a set of both natural and synthetic images.