In the present study, the groups of cutting conditions that minimize surface roughness and its variability are determined, in ball-end milling operations. Design of experiments is used to define experimental tests performed. Semi-cylindrical specimens are employed in order to study surfaces with different slopes. Roughness was measured at different slopes, corresponding to inclination angles of 15°, 45°, 75°, 90°, 105°, 135° and 165° for both climb and conventional milling. By means of regression analysis, second order models are obtained for average roughness Ra and total height of profile Rt for both climb and conventional milling. Considered variables were axial depth of cut ap, radial depth of cut ae, feed per tooth fz, cutting speed vc, and inclination angle Ang. The parameter ae was the most significant parameter for both Ra and Rt in regression models. Artificial neural networks (ANN) are used to obtain models for both Ra and Rt as a function of the same variables. ANN models provided high correlation values. Finally, the optimal machining strategy is selected from the experimental results of both average and standard deviation of roughness. As a general trend, climb milling is recommended in descendant trajectories and conventional milling is recommended in ascendant trajectories. This study will allow the selection of appropriate cutting conditions and machining strategies in the ball-end milling process.