Abstract: The nonlinear frequency modulation signal(NLFM) is widely used in the field of radar and sonar. It has the characteristics of multiple order polynomial phase and many unknown parameters, which makes parameter estimation difficult. According to this, a method of Parameter estimation of nonlinear frequency modulated signal under the sparse reconstruction is presented. The method utilizes good time-frequency characteristics of Gabor atoms, uses norm Sparse regular least squares model and deduces the Second Order Cone Programming (SOCP) form of the problem. Finally it estimates parameters with Gabor atoms screened out. The method analyzes the time-frequency characteristics of the signal, completes the decomposition and reconstruction of the signal and suits all kinds of FM signals. Simulation results show that the precision to second order and first order phase parameter of the proposed method is closed to CRB, and is more suitable to low signal-to-noise ratio while compared with second phase difference algorithm.