In order to expand the capability of Compressed Sensing (CS), a fusion method for multi-focus image using Dual-tree Contourlet (DT-Contourlet) and CS is proposed. First, the source images are decomposed using DT-Contourlet for extracting multiscale and direction information while overcoming the limitation of traditional contourlet which is lack of shift invariance. Then, in DT-Contourlet domain, the decomposition coefficients are treated as containing two components, i.e., dense and sparse components. The dense components are fused using selection method by introducing neighborhood gradient as clarity index to indicate the characteristics of defocus. The sparse components are fused under the framework of CS via fussing a few linear measurements by solving the problem of L1 norm minimization which is based on a two-step iterative shrinkage/threshold reconstruction algorithm. The experiments demonstrate that the convergence rate of reconstruction is faster than that of orthogonal matching pursuit. Meanwhile, the proposed method provides more satisfactory fusion results in terms of visual quality and quantitative criterion.