Probabilistic load flow (PLF) calculation is an important tool for power system steady state performance analysis. Most of the existing PLF methods require to know the probability distribution functions of input random variables, which are very difficult to be modeled accurately. In the paper, a PLF method based on gravitational search algorithm(GSA) and Latin hypercube sampling (LHS) was proposed. It has many advantages, such as high accuracy and fast computation. It is unconstraint by the type of the input random variables probability distributions. Sequential quadratic programming (SQP) algorithm is incorporated in GSA as a local search mechanism to improve the performance of the conventional GSA algorithm, The validity of proposed method was tested on IEEE 14 bus system and IEEE 118 bus system. The simulation results show that the proposed method is effective, which not only has the advantages of fast computation, high accuracy and good robustness, but also can deal with the correlation between input random variables flexibly. The proposed method is of some practical engineering value.