This paper describes errors and corrections in snow heat flux when it is calculated using the numerical differentiation and integration method. The data obtained by the 20th Japanese Antarctic Research Expedition during GARP-POLEX in 1979 are used as a test case. Four factors are considered as causes of errors: a) temperature resolution, b) integration, c) determination of snow density and d) the deepest boundary condition. Factors a) and b) are significant in short term estimation, and the total error exceeds 90% if a daily value is calculated. The errors from a) can be reduced if the temperature is averaged over a long period, while those from b) become small in long term flux calculations. The total error can be reduced to 10% in monthly flux, while the improvement is limited by c) and d). If a constant thermal diffusivity is assumed between two levels of temperature measurement, a numerical filter which compensates for the effect of b) is composed. By using running averages in deep layers (z > 0.5 m) and the numerical filter in shallow layers (z <0.5 m), hourly flux can be calculated with errors of about 30%.