When applying Grover's algorithm to an unordered database,the probability of obtaining correct results usually decreases as the quantity of target increases.A four-phase improvement of Grover's algorithm is proposed to fix the deficiency,and the unitary and the phase-matching condition are also proposed.With this improved scheme,when the proportion of target is over 1/3,the probability of obtaining correct results is greater than 97.82%with only one iteration using two phases.When the computational complexity is O((√)M/N),the algorithm can succeed with a probability no less than 99.63%.