The choice of coordinate system to calculate eddy covariance fluxes becomes particularly relevant at complex measurement sites. The traditional way is to perform double rotation (DR) of the coordinate system i.e., to calculate turbulent fluxes in a coordinate system that is aligned with the flow streamlines within the flux averaging period (e.g., Kaimal and Finnigan, 1994). The second approach, the so-called planar-fitted (PF) coordinate system, averages the flow over a longer period of time, in practice a month or more. The PF method allows to derive an intercept coefficient of the vertical wind speed which can be attributed to the offset of the sonic anemometer or the average vertical flow related to meteorological conditions. We evaluated the variants of the PF methods using data from a variety of sites ranging from complex urban and forest sites to nearly ideal forest and peatland sites. At complex sites, we found that the intercept of the vertical wind speed derived from the PF method is a function of wind direction, time of day and/or stability. The sector-wise PF (SPF) method frequently led to insignificant statistical relationships. We tested a continuous PF (CPF) method where the relationship establishing the coordinate frame was represented as the continuous function in the form of Fourier series. The method enabled to obtain the PF with lower uncertainty as compared to the SPF method, by selecting necessary number of harmonics for each site based on confidence intervals of estimated parameters. Therefore, we recommend to use the CPF method in cases when the number of observations in some wind direction interval is low or the obtained SPF is insignificant due to large variance in measurements. We also showed that significant systematic difference can exist in cumulative turbulent fluxes between the DR and PF methods over a longer period of time. Derived vertical advection of carbon dioxide exhibited large variability with wind direction due to topography at complex sites and therefore, without considering horizontal advection, cannot be used to improve the net ecosystem exchange estimation during nocturnal, low turbulence conditions.