Without reliable trust relationship established between cooperative parties, the paradigm of largescale resource sharing and cooperative problem solving as envisioned by most will not come true in Grid. What's more, for the scale and openness of Grid environment, cooperation often occurs among completely unknown entities, where traditional identity-based trust system obviously cannot work well for its restricted scalability and flexibility. In this paper, we propose a pre-evaluating set based trust model, which resorts to reputation mechanisms for trust establishment. The introduction of the preevaluating set is to track an entity's rating criteria, overcome the prevalent strangeness of Grid entities. With this pre-evaluating set, we can reasonably filter out malicious ratings by examining inconsistency between ratings to this set and to real transactions, effectively tune a rater's bias to cater to the current evaluator's criteria by interpolation approach and scientifically weight a rater's rating in aggregation by the introduction of Criteria Coherent Degree which describes the similarity between an entity?s rating criteria and the current evaluator?s.