Motif discovery is a significant problem in Bioinformatics. According to the complexity of most signals in biologic sequences, there are no extremely good models or dependable algorithms to solve this problem. This paper introduces the Uniform Projection and Neighbourhood Thresholding (UPNT) algorithm, which is based on two efficient strategies: Uniform Projection and Neighbourhood-based Thresholding. In the UPNT algorithm, the policy of uniform projection leads to fewer projections, while the strategy of refining the buckets after aggregation results in great abatement of the number of buckets to be refined. This paper further demonstrates its superiority over other projection algorithms by the experiments.