Data from 2435 gas-liquid flow experiments in horizontal pipelines, taken from different sources, including new data for heavy oil are compiled and processed for power law and composite power law friction factor correlations. To our knowledge this is the largest database so far published in the literature; it includes the widest range of operational conditions and fluid properties for two-phase friction factor correlations. Separate power laws for laminar and turbulent flows are obtained for all flows in the database and also for flows sorted by flow pattern. Composite analytical expressions for the friction factor covering both laminar and turbulent flows are obtained by fitting the transition region between laminar and turbulent flow with logistic dose curves. Logistic dose curves lead to rational fractions of power laws which reduce to the power laws for laminar flow when the Reynolds number is low and to turbulent flow when the Reynolds number is large. The Reynolds number appropriate for gas-liquid flows in horizontal pipes is based on the mixture velocity and the liquid kinematic viscosity. The definition of the Fanning friction factor for gas-liquid flow used in this study is based on the mixture velocity and density. Error estimates for the predicted vs. measured friction factor together with standard deviation for each correlation are presented. The correlations in this study are compared with previous correlations, homogeneous models and mechanistic models most commonly used for gas-liquid flow in pipelines. Since different authors use different definitions for friction factors and Reynolds numbers, comparisons of the predicted pressure drop for each and every data point in the database are presented. Our correlations predict the pressure drop with much greater accuracy than those presented by previous authors. (C) 2003 Elsevier Ltd. All rights reserved.