Experimental results on evapotranspiration (ET), relevant to small on-site facilities are presented, derived from one-year controlled experiments in five pilot-scale horizontal subsurface flow (HSF) constructed wetlands (CW) used as lysimeters. The CW units operated in Northern Greece. They were rectangular tanks made of steel, with dimensions 3m long, 0.75m wide and 1m deep. Three different porous media were used, i.e., medium gravel, fine gravel and cobbles. Two plants were used, namely common reed (R, Phragmites australis) and cattails (C, Typha latifolia). One unit was unplanted. ET was estimated based on the water budget method. Conclusions were drawn on its relation to season and vegetation density. Furthermore, Pearson correlation coefficient analysis identified the main factors affecting wetland plant ET. Seven well-known ET empirical methods were applied to estimate ET using the measured meteorological and wetland data. ET estimated by the empirical methods were multiplied with appropriate correction coefficients to match measured ET, providing this way appropriate plant coefficient (K(c)) values, and equations for predicting HSF CW evapotranspiration. The suitability of these methods for the particular constructed wetland type is discussed through comparison with the measured data. The Blaney-Criddle method was found as best. Furthermore, stepwise multiple linear regression analysis was used with the measured ET and meteorological data to produce simple empirical equations to predict ET rates according to meteorological factors, plant and substrate material.