For the scale, openness and sharing of grid environment, how to establish a reliable trust relationship between unknown entities more and more challenges grid systems. With wide application in electronic commerce and online communities, reputation mechanism emerges as a promising solution, where a scientific reputation evaluation method is crucial. Yet, most methods currently available do not consider grid's distinct characteristics such as the sparseness of ratings and strangeness of entities, which are not feasible to grid. In this paper, we propose a pre-evaluating set based method for reputation evaluation in grid. The introduction of the pre-evaluating set is to overcome the sparseness and strangeness mentioned above. With this set, we can effectively tune a rater's bias to cater to the current evaluator's criteria and reasonably weight a rater's rating in aggregation. For bias-tuning, we flexibly turn this into an interpolation problem and conveniently solve it in Matlab. For the final aggregation, we introduce the notion of criteria coherent degree for weight calculation