The invention provides a big data parallel operation-based voice mail business data analysis method. The method comprises the following steps: collecting structural logs thrown by core components of a voice mail platform business system into a Hadoop offline cluster through a flume log collection system; generating or updating user model data from original logs collected the same day through MapReduce; updating the logs collected the same day through the user model data so as to generate corresponding pipeline model data; and querying the latest user model data or pipeline model data in a statistical time interval to obtain a statistical analysis result. During the business statistical analysis, the condition of processing the whole data set during one query is avoided, so that the parallel operation ability of a big data analysis framework is effectively utilized, and the reckless action is skillfully avoided. Through the method provided by the invention, the query time is decreased, and the data processing efficiency is improved; and the method is suitable for the voice mail systems with massive users and business data.