In the early stages of transmission system development, the selection of the most suitable gearbox layout is a critical task because design concepts are typically not fully defined and physical prototypes are unavailable. For this reason, simplified calculation models are often employed to support the design process. The parameters used in such models inherently depend on the assumptions and approximations adopted and are therefore unavoidably affected by epistemic uncertainty. This work presents a methodology for evaluating alternative gearbox configurations by integrating analytical modeling of power losses with uncertainty quantification through fuzzy arithmetic. The focus is placed on sliding and windage losses, which are estimated using analytical models incorporating key geometric and operating parameters. Recognizing the uncertainty inherent in these parameters, the method propagates their variability through the models to generate efficiency intervals, rather than single-point estimates. The proposed framework considers both the nominal efficiency at the design operating point and the sensitivity of each layout to input variability. A case study compares three common architectures (single-stage, double-stage, and planetary transmissions) under equivalent boundary conditions of transmission ratio and packaging. Results highlight how the assessment of uncertainty bands can help with more informed performance rankings, emphasizing the importance of including these considerations in early design phases. This framework could lead towards more robust concept selection, while remaining compatible with progressive model refinement based on experimental data in later development stages.