Abstract: The low calculation efficiency of the genetic algorithm (GA) method is an obstacle to 3D resistivity inversion. Moreover, some improved methods which are time-consuming but beneficial for the inversion effect and the search efficiency can not be used in GA due to their low calculation efficiencies. To solve the above problems, a multi-level master-slave parallel computing strategy for GA is put forward based on the natural characteristics of parallel computing. Through this improvement, a generating method for strictly uniform initial population is proposed, with which the initial generation can be closer to the optimal solution. A random-ratio arithmetical crossover algorithm is proposed based on the differences of fitness values between the cross-individuals, which can keep genetic competition advantages of the better individual. Then the joint mutation algorithm is presented, which is the combination of the traditional random mutation algorithm and the deterministic search optimization algorithm in the linear inversion. It can maintain the randomness of the mutation and optimize the mutation direction. Eventually a 3D resistivity inversion using an improved parallelized GA is formed. The performance of the improved parallel GA is evaluated in synthetic and practical cases. The examples illustrate that the improved parallel GA can enhance the calculation efficiency significantly and has obvious advantages in searching the optimal solution, suppressing the false anomaly and obtaining high-quality inversion results. The improved parallel GA provides an effective way for 3D resistivity inversion imaging in practical projects.