This research evaluated the performance of fourteen stem taper models for their use in Eucalyptus tereticornis forest plantations. Three types of models were evaluated: polynomials, segmented and variable-exponent models. A sample of 130 trees collected in forest plantations with stand ages between five and twelve years-old located at the Atlantic Colombian coast was used. During the parameter estimation process a continuous time autoregressive model (CAR) was used in order to expand the error structure, reducing the effect of the autocorrelation generated by the longitudinal measurements taken along the stem. The model evaluation considered a comparison of the predictive ability respect to the prediction of upper-stem diameter, upper-stem heights, merchantable volume until a given stem-height and merchantable volume until a given stem-diameter. Measures of bias, precision and error were used for evaluating the models. A third-order continuous-time autoregressive model CAR(3) allowed to adequately model the residual autocorrelation for all stem profile models. The variable-exponent models showed the best performance in terms of goodness of fit and predictive capability.