Spreader classification is an online traffic measurement function that has many important applications. In order to keep up with ever-higher line speed, the recent research trend is to implement such functions in fast but small on-die SRAM. However, the mismatch between the huge amount of Internet traffic to be monitored and limited on-die memory space presents a significant technical challenge. In this paper, we propose an Efficient Spreader Classification (ESC) scheme based on dynamic bit sharing, a compact information storage method. We design a maximum likelihood estimation method to extract per-source information from the compact storage and determine the heavy spreaders. Our new scheme ensures that false positive/negative ratios are bounded. Moreover, given an arbitrary set of bounds, we develop a systematic approach to determine the optimal system parameters that minimize the amount of memory needed to meet the bounds. Experiments based on a real Internet traffic trace demonstrate that the proposed spreader classification scheme reduces memory consumption by 3–20 times when compared to the best existing work. We also investigate a new multi-objective spreader classification problem and extend our classification scheme to solve it.