Most of the existing algorithms for mining frequent items over data streams do not emphasis the importance of the more recent data items. We present an efficient algorithm where a fading factor lambda is used for computing frequency counts exceeding a user-specified threshold over data streams. Our algorithm lambda-Miner can detect epsilon-approximate frequent items of a data stream using O(epsilon−1) memory space and the processing time for each data item is O(1). Experimental results on several artificial data sets and real data sets show that lambda-Miner performs better than lambda-LC in terms with precision, memory requirement and time cost.