The ability of the balance drum to balance the axial force is the key factor for the failure of multistage centrifugal pump. However, during the operation of the balance drum, due to the long-term collision with the liquid or the friction with the casing, the leakage amount at the balance drum clearance is gradually increased, resulting in the balance drum being worn. Therefore, studying the leakage flow is critical to the ability of the balance drum to accurately balance the axial force of the entire impeller. In this paper, three-dimensional turbulent flow of multistage centrifugal pump was simulated by using the CFD code FLUENT. Besides, steady simulation was conducted for different operating points of the pump, the turbulence was simulated with shear stress transportation(SST) turbulence model together with automatic near wall treatment. CFD results were compared with those from the model test. And the results of the pressure and leakage in the balance pipe and the external characteristics of the multistage pump were basically consistent with the experimental results. Moreover, the maximum errors of head, efficiency and shaft power were 4.17%, 2.81% and 4.25% respectively, but the experimental flow rate of balance pipe was always greater than the simulated one. This was mainly because the influence of orifice flowmeter in the pipe had been not taken into account in the numerical simulation. The maximum error of the flow rate of the balanced pipe at the design point was 4.49%. The maximum error of pressure was 2.5%. It showed that the calculation method selected in this paper could provide a reliable guarantee for this study. The results showed that at the design flow rate, the liquid pressure in the front cavity of the first impeller increased gradually along the axial direction from the inlet section to the outlet section. When the balance drum clearance was less than 0.2 mm, the pressure distribution along the radial direction was uniform in each section. But when the clearance was more than 0.2 mm, it was asymmetric. Furthermore, with the increase of clearance, the pressure inhomogeneity became more obvious. Moreover, at 0.5Q(Q is design flow, Q=128 m3/h) and 1.5Q flow rates, there was no obvious regularity of pressure increment along the radial direction under different clearances. The pressure increment was the smallest when the balance drum clearance was 0.3 mm, and the biggest when the balance drum clearance was 0. And under the above 2 conditions, the minimum increments were 50.7% and 88.9% of the maximum, respectively. When the clearance increased from 0 to 0.5 mm, under design flow rate, the pressure increased gradually along the radial direction. Wherein, when the clearance was 0 and 0.5 mm, the pressure increment was the maximum value and the minimum value, respectively, and the minimum value was 44.6% of the maximum value. Besides, there was a large vortex region in the front cavity of the first impeller, when the clearance of balance drums was 0, 0.3 and 0.5 mm, respectively. The vortex region decreased gradually with the increase of flow rate, which indicated that the appearance of the vortex region was the main reason for the change of pressure in the cavity. In addition, with the increase of clearance, the total axial force of 11 stage impellers decreased first and then increased. And when non-dimensionalized balance drum clearance area was greater than 6.6×10-3, the balance drum could not effectively balance the axial force. Furthermore, the bearing had a greater risk of fracture in this range. This research can provide useful reference for design of balance drum and prediction of risk of bearing fracture.