An approximate method to determine sample size for the estimation of population variance, σ(2), is given. The estimate of σ(2) is denoted as s(2) . Based on the assumption of a normal distribution for (s(2)/σ(2)-1), the sample size is approximately equal to 20,000 z(2) p,/k(2); where z is a standard normal deviate, p is the probability that Δs(2) (≡ 100¦s(2) - σ(2)¦/σ(2)) is less than, or equal to, a critical value k, and k (measured as gDs(2)) is the desired precision of s(2) .The expected value of Δs(2), with respect to sample size, and the expected cumulative frequencies of Δs(2) over sample size for various k values are given. Their goodness of fit to the observed results was satisfactory except for populations that were different from normal. The observed values were taken from a study on four yield components in five sugarcane polycross progenies, grown in two contrasting environments over 2 years in three selection stages.The expected Δs(2) was found to be independent of the population coefficient of variance.