The sparsity of a star image was explored in different representation approaches to apply better sparse representation to the compressive imaging process of a star tracker. The sparsity of star image was analyzed in two ways. In the first way, the Discrete Cosine Transform (DCT), Discrete Fourier Transform (DFT) and the Discrete Wavelet Transform (DWT) were used to construct complete representation bases and to examine the sparsity of star image in different complete representation bases. In the second way, the DCT complete basis was selected to create the overcomplete dictionary and learning dictionary to analyze the sparsity of the star image in different representation approaches. The simulation result shows that the average Peak Signal to Noise Ratio(PSNR) of the star image is 15-20 dB higher than that of the common scene image by complete based representation, while the overcomplete dictionary and learning dictionary based representations improve the PSNR by 2-20 dB with different sparsities. Regarding the quality of star point reconstruction, the rate of successful reconstruction is mostly higher than 95% when the sparsity is more than 10% in different representation approaches. The results verify that the star image has the preferable sparsity and meets the requirement of compressive imaging.The reconstruction of the star image maintains the position of star centroid position suitable for attitude determination to a large extent, which verifies the sparse precondition and feasibility for applying the compressive sensing in star trackers.