A new least-squares method for inversion of d.c. resistivity soundings from horizontally layered earth is proposed. Powers are used as measured data. Apparent resistivity is generalized and apparent parameters curves are defined. Estimates for parameter errors are often possible even when data errors are unknown or large. Acceptance or rejection of a model is made an easier decision. Total computational effort is often reduced when compared to global least-squares methods, because although several optimizations are needed, few data points are involved in each, and initial guesses are requested only once and not all at the same time. Correlations between parameters may be used to improve solutions. Precise indications for extra needed data points can be conveyed. A test example is worked out and a field sounding, taken by Rijo (1973, personal communication) in an alluvium-filled river valley in north-eastern Brazil, is studied.