Abstract: A new track initial algorithm based on layered clustering method is proposed to solve multi-target detection problem with passive reconnaissance data. Especially when the passive reconnaissance scanned aperiodically, the capture plots are fragmentary, and the prior information of targets number and athletic characteristics are insufficient. The algorithm effectively utilized the attributive characteristics to solve the track initial problem. Firstly, the observation set was rough clustered according to the systems of Pulse Frequency (PF), Pulse Recurrence Frequency (PRF), Pulse Width (PW) electromagnetic parameter; Secondly, the exact result of classification was get by clustering of electromagnetic parameter using K-means algorithm; Thirdly, computing the velocity of each dimension of all of the probable point pairs to eliminate the illusive observations by space-time constrained conditions; Ultimately, the final initialed track can be achieved by extended search approach. Experiments on both simulated and real world data showed its effectiveness and practicability.