Accurate evaluation of evapotranspiration (ET) flux is an important issue in sustainable urban drainage systems that target not only flow rate limitations, but also aim at the restoration of natural water balances. This is especially true in context where infiltration possibilities are limited. However, its assessment suffers from insufficient understanding. In this study, ET in 1 m3 pilot rain gardens were studied from eight lysimeters monitored for three years in Paris (France). Daily ET was calculated for each lysimeter based on a mass balance approach and the related uncertainties were assessed at ±0.42 to 0.58 mm. Results showed that for these lysimeters, ET is the major term in water budget (61 to 90% of the precipitations) with maximum values reaching 8–12 mm. Furthermore, the major determinants of ET are the existence or not of an internal water storage and the atmospheric factors. The vegetation type is a secondary determinant, with little difference between herbaceous and shrub configurations, maximum ET for spontaneous vegetation, and minimal values when vegetation was regularly removed. Shading of lysimeters by surroundings buildings is also important, leading to lower values. Finally, ET of lysimeters is higher than tested reference values (evaporimeter, FAO-56, and local Météo-France equations).