Three models for the prediction of bamboo culm length and two for culm volume were fitted from data of 303 guadua bamboo (Guadua angustifolia Kunth) culms. Data are from 101 temporary inventory plots systematically distributed over the coffee region of Colombia (inventory area about 1,029,525 ha). Linear and non-linear regression models were used, and the precision of the models was evaluated by cross-validation. The models were compared by studying the adjusted coefficient of determination, the bias, mean square error and Akaike's information criterion and by the F-test that compares predicted and observed values. For culm length, the best fit showed models that included predictor variables related to stand characteristics such as quadratic mean diameter and number of culms per ha. For culm volume prediction, the inclusion of culm length improved the model significantly. For the simple form factor model, precision of prediction was least. The models developed are useful in facilitating the estimation of stand characteristics that are relevant for the silvicultural management of Guadua stands and also for the assessment of their environmental services (such as carbon sequestration).