Transmission error is widely regarded as one of the primary internal sources of excitation in geared transmissions, leading to vibration and noise. A comprehensive literature review conducted in this dissertation showed that machining errors of gear teeth have a significant influence on both the magnitude and the waveform of transmission error. Nevertheless, many studies still assume ideal tooth geometry or limit the analysis to intentional micro-geometric modifications only. Therefore, the present work investigates the influence of machining deviations on the static transmission error of involute cylindrical gears, with deviations represented either by stochastically generated deviation fields corresponding to different tolerance classes and statistical distributions, or by deviation fields derived from scanned tooth-flank topography. For the latter, their interaction with tip-relief modifications of different parameters is also examined. The effects of realistic tooth-flank geometry on contact pressure and tooth-root stress are evaluated as well. Particular attention is given to the influence of pitch error, inherently included in the scanned geometry, on static transmission error, contact pressure, and tooth-root stress. To enable systematic and repeatable numerical analysis of gear engagement, a parametric 2D and 3D finite element model was developed. The developed macro, written in APDL (ANSYS Parametric Design Language), automatically generates the gear-pair geometry and sets up the finite element contact analysis based on user-defined parameters, including tooth-flank deviation fields and tip-relief modification parameters. The simulation results obtained for spur gears are complemented by experimental measurements of quasi-static transmission error performed on a closed-loop gear test rig for both spur and helical gears. The test rig, designed primarily for closed-loop operation, also allows operation in an open-loop configuration only under no-load conditions; quasi-static transmission error was measured in both configurations. The experimental results indicate that the reaction gearbox does not significantly influence the transmission error of the tested gear pair in the closed-loop configuration. Furthermore, incorporating scan-derived deviation fields into the computational model can yield correlation coefficients of up to 0.98 between simulated and measured transmission error waveforms. The results also show that tip relief remains effective when applied to realistic tooth-flank geometry, although it is less effective than in the case of ideal geometry. Overall, the results demonstrate the importance of considering realistic tooth-flank deviations in the computational prediction of gear behaviour, since idealized geometry may lead to an incomplete representation and significant underestimation of transmission error, including its peak-to-peak values, and stress-related quantities.