A multiplicative spread spectrum watermarking technique in curvelet domain is presented. Watermarked curvelet coefficients are modeled using generalized Gaussian distribution, Laplacian distribution and Cauchy model. Watermarking detectors are designed employing locally most powerful (LMP) approach. The detection performances of three detectors are analyzed. Experimental results show LMP detector based Cauchy model is superior to the detectors based generalized Gaussian and Laplacian distribution.