Abstract Since the launch of the gravity recovery and climate experiment (GRACE) satellites in March 2002, data quality has been improved significantly. For the star camera and accelerometer data, the calibration parameters have been well determined and used in data generating and processing. The purpose of this paper is to investigate how well the GRACE satellite orbits can be determined using improved accelerometer data and to assess the differences between the use of accelerometer data and the use of a priori models for the surface forces. As is known, the orbit accuracy depends on the force models used in the dynamic orbit determination, but the models for the surface forces acting on low-Earth satellites are uncertain. To alleviate this problem, the GRACE concept uses a three-axis accelerometer to measure the non-gravitational accelerations. To reduce the effects of force model error on precise orbit determination, one can estimate a set of empirical parameters. In the case of highly variable forces acting on satellites (such as high solar activity), the accelerometer provides high quality measurements of the phenomena. For this investigation, we have analyzed the data during high solar activity. The orbit accuracy is assessed using a number of tests, which include analysis of orbit fits, Satellite Laser Ranging (SLR) residuals, K-Band Ranging (KBR) residuals and external orbit comparison. The results show that an accuracy of about 2 cm in each component is achieved for the GRACE orbits and better orbits can be determined using accelerometer data during high solar activity. In addition, the GRACE orbits generated with the accelerometer data are generally better in the high frequency range.