The measured pressure drop of R134a, flowing downward and horizontally inside smooth and corrugated copper tubes, is estimated by the closed form of artificial neural network method to have a reliable empirical correlation using some dimensionless numbers. The working fluids are R134a and water flowing in the test tube and annular tube, respectively. This paper is a continuation of the authors' previous work and includes all their previous works about condensation and boiling in tubes. All data used in the present paper are obtained from the authors' previous studies. The training sets have the experimental data of convective condensation and boiling experiments including various mass fluxes and saturation temperatures of R134a. Froude number, Weber number, Bond number, Lockhart and Martinelli number, void fraction, the ratio of density to dynamic viscosity, liquid, vapor and equivalent Reynolds numbers, surface tension parameter and liquid Prandtl number are the inputs of the formula as the dimensionless numbers obtained from measured values of test section, while the output of the formula is the measured pressure drops in the analysis. A closed form of multi-layer perceptron (MLP) method of artificial neural network (ANN) is used to estimate the experimental pressure drop of R134a numerically. 1177 data points are used in the analyses of the ANN method to be able to have a single generalized empirical correlation for both condensation and boiling flows. The evaluation of the closed form of multi-layer perceptron (MLP) with two or three inputs and one hidden neuron architecture was successful predicting the measured pressure drops with their error bands being within the range of +/- 30% for all used data. The proposition of empirical correlations are performed for both condensation and boiling flows separately. A single empirical correlation is able to calculate the measured pressure drop of both condensation and boiling flows together. Moreover, the dependency of output of the proposed formula from input values is examined in the study. By means of the dependency analyses, liquid Prandtl number, Butterworth's void fraction and Lockhart and Martinelli parameter are found to be the most dominant parameters among other dimensionless numbers. (C) 2013 Elsevier Ltd. All rights reserved.