In carbonate reservoirs, the relationship between porosity Ø, a measure of the combined volumes of several kinds of pore space (e.g. interparticle and separate-vug), and permeability K are neither linear nor logarithmic, hence only weakly correlatable. Approaches to an estimation of permeability that employ both petrographical and petrophysical parameters, the so-called rock-typing techniques, have proven to be the most nearly precise. However in many studies simple K/Ø cross-plots are used for each rock-type to provide trendlines from which K values are derived as a function of Ø values; this is common practice even though the coefficient of correlation r2 departs significantly from 1. This paper describes and provides examples of an improved technique of rock-type classification in which each rock-type is characterized by a discrete and unique Gaussian distribution of log K. It facilitates upscaling for it suggests the use of a single geometric mean value for permeability and a corresponding standard deviation (variance, or coefficient of variation) for each rock-type. This new technique can be used in uncored wells by extrapolating these determinations into wireline logs as documented below in a case study of layer-cake reservoirs in a field of the Abu Dhabi offshore (U.A.E.).