Cylindrical models, although scarcely used, are helpful tools in atmospheric science applications such as siting wind machines or setting up atmospheric pollution monitors. In this paper, three simple models have been successfully compared. The first, a second-order model, is an addition of two harmonic functions and was obtained by a multiple linear regression. It may be expressed in closed form and for this reason was taken as the reference model. The second was a weighted local average of directional neighbours and the third a robust variant of the second model. These models were used with wind speed and temperature data obtained from a RASS sodar during April 2001. The measurements were 10-min averages calculated at 20 m intervals from 40 to 500 m. Two different analyses were considered, namely, daily evolution and directional behaviour. In the first, there was a good agreement between the three models. Major differences were observed in the transitions between day and night at intermediate levels, although they were smoother in the second-order model. The directional analyses showed that the prevailing directions were more poorly defined by the second-order model, although it may be considered a reasonable approximation. The weighted local average model presented oscillations, which is not desirable from a practical point of view; the third model was more robust. As a result, the second-order model was preferred in daily analyses of lower levels due to its closed form. The coefficients of this model for daily evolution were successfully parameterised as a function of height. Finally, the undesirable oscillations of the second model disappeared when the weighting interval widened, though prevailing directions were less well defined. (c) 2006 Elsevier Ltd. All rights reserved.