In the GRAPES-3 experiment, consisting of an array of $\sim$400 scintillator detectors, the arrival direction of the shower is determined from the relative arrival times of particles at different detectors.The fixed arrival time of the signal from the detector to the measuring device, referred as time offset, is crucial for an accurate measurement of shower direction. In the older method the time offset of various detectors was measured with respect to a common detector. But this method proved to be ineffective since it took a long time ($\sim$40 days) to complete one round of measurements. However, the time offsets vary with temperature due to change in the propagation delay in signal cables.Hence, a technique was devised to determine the time offsets on an hourly basis by using the shower data. In this method, the time offset between two neighbouring detectors was determined from the distribution of their relative arrival times. A random walk method was used to effectively determine the time offset with respect to a common detector. The accuracy of the new method was validated by using both the simulations, and EAS data which will be presented at the conference.