The present study uses data obtained during the Terrain-induced Rotor EXperiment (T-REX), conducted during March and April of 2006. Two 34-meter NCAR ISFF towers were positioned along the Owens Valley in California, while a third tower was positioned on the east facing slope of the Sierra Nevada mountain range. Each tower was equipped with CSAT3 ultrasonic anemometers on six levels (5, 10, 15, 20, 25 and 30 m), capturing 3D wind speed and sonic temperature with a sampling frequency of 60 Hz. A main feature of the data preprocessing was the application of five stationarity criteria, providing us with differently defined stationary data, which are necessary in order to properly apply MOST. This study focuses on investigating the impact of the stationarity criteria on the form of the flux-variance similarity relationships in MOST. Since MOST applies only in the constant flux layer, we first assess the degree of vertical divergence of the momentum and heat fluxes. From our calculations, we conclude that local similarity scaling, rather than the usual surface layer scaling, should be used in this study. We calculate the dimensionless standard deviations of wind velocity components and potential temperature as a function of the stability parameter z/L and find that the use of data selected using the five different stationarity criteria results in substantial changes in the form of the similarity functions. The free coefficients α and β in the flux-variance similarity relationships, calculated with the least-squares method, are within the range of those obtained in other studies. However, the near-neutral asymptotic values of α for the dimensionless standard deviations of longitudinal and lateral wind speed components are somewhat larger compared to usual, textbook flat terrain values. We argue that this observation may be due to larger eddies having good memory of the upwind terrain conditions. On the other hand, the value of α for the dimensionless standard deviations of the vertical wind speed component are smaller than typical flat terrain values. Consequently, an anisotropy of the flow, reflected in the α values of the similarity functions, is more pronounced in our data compared to flat terrain studies. Results from a self-correlation analysis indicate potential issues with the local scaling approach for the horizontal wind components, especially for stable conditions. Finally we assessed the reduction in the scatter around the best-fit similarity functions as a function of the applied stationarity criteria. We find that on average, the scatter is most reduced during daytime, unstable conditions, especially in the case of using stationarity criteria that are based on detecting intermittency of the flow.