This paper utilizes a BP neural network based on Bayesian Algorithm to quantify the possible horizontal groove defects on tank floor of oil and gas from Magnetic flux leakage (MFL) signals. This method corporates the Bayesian Algorithm into the BP neural network model to control the complexity of the model and optimize weights of the network in order to build the relationship between the MFL signals and the defect features, which can save the training time of the network and accurately quantify the defect profile. In order to obtain more defects information, this paper uses the three-axial MFL signals to quantify the horizontal groove defects, which further improve the accuracy of quantification. The simulation results show that the method proposed in this paper exhibits better performance in both efficiency and accuracy for the quantification of horizontal groove defects.