Paleomagnetists dealing with deep drill samples cannot use the classical Fisher statistics when they don't know the declinations of the magnetization of their samples. The arithmetical mean of the inclinations gives a strongly biased estimation of the true mean inclination when these inclinations are steep and scattered. By fitting cumulative distribution curves, we show that we can obtain an unbiased (or at least less biased) estimation of the true mean inclination and of the k value. The method is tested on artificial and real cases. ¿ 1998 American Geophysical Union