The accuracy of the live-time circuit of a 400-channel analyzer was studied in detail, and was found to be unsatisfactory even for long-lived radionuclides. It was found that automatic live-time correction with the multi-channel analyzer gave rise to increasing positive errors with increasing count rate; this overall positive error was composed of a positive error due to the slowness of the electronic circuitry, and a smaller negative error due to the finite pulse-width. Adequate correction could be performed by feeding the information from the dead-time output of the multi-channel analyzer to an external live-time circuit with variable oscillator frequency and pulse-width. Four methods for dead-time correction were compared experimentally in the case of short-lived radionuclides (T as low as 7 sec): the method of Bartošek et al., the method of Schonfeld, the use of a sufficiently short counting time as compared to the half-life, and the live-time mode of counting without additional correction. These four methods were applied to the determination of oxygen and silicon in rocks by 14 MeV neutron activation analysis. Results are given for USGS standard rock G-2.