Compressed sensing seeks to recover a sparse or compressible signal from a small number of linear and non-adaptive measurements. Gaussian random matrix is a kind of fundamental measurement matrices, but its performance isn’t perfect because of more errors in recovery. This paper studies a new kind of matrix based on improving Gaussian random matrices. Measure sparse signals with improved matrices and recover original signals with orthogonal matching pursuit. Numerical experiments showed that the quality of recovered signal by improved measurement matrices is better than Gaussian random matrices.