We study theoretically and experimentally the quantification of non-gaussian distributions via nondestructive measurements. Using the theory of cumulants, their unbiased estimators, and the uncertainties of these estimators, we describe a quantification which is simultaneously efficient, unbiased by measurement noise, and suitable for hypothesis tests, e.g., to detect nonclassical states. The theory is applied to cold 87Rb spin ensembles prepared in non-gaussian states by optical pumping and measured by nondestructive Faraday rotation probing. We find an optimal use of measurement resources under realistic conditions, e.g., in atomic ensemble quantum memories.