on-axis Fourier digital holography, the conventional way for image reconstruction is inverse fast Fourier transform (IFFT), but the Shannon′s sampling theory must be satisfied in the sampling process, which causes the data redundancy and greatly increases the cost of storage and transmission. A reconstruction method of phase-shifting on-axis Fourier digital holography based on compressed sensing (CS) theory is put forward. Firstly, on-axis hologram is collected in the Mach-Zehnder interferometer light path. Then the collected data are partially sampled and measured. Lastly the image is numerically reconstructed by using the minimum total variation method. The numerical simulation result shows that the reconstruction method based on CS, which takes the sampling and compression of the hologram data to one step, is superior to that based on the conventional reconstruction method. The amount of sampling data based on CS is less than that based on IFFT, but the original image can also be accurately reconstructed from just about 8% hologram data.