Understanding the interaction between the hydrological cycle and native forests is essential to improve watershed management and support ecological services. The objectives of this study were to analyse and evaluate rainfall interception models applied to an ecosystem from the Brazilian Atlantic Forest Biome (BAFB), classified as “Montane Semi-Deciduous Forest.” The variables Gross Rainfall (P) (rainfall that reaches the canopy), Throughfall (Tf), Stemflow (Sf), and Interception Loss (I) were measured from September 2012 to March 2015. Weather variables were quantified above the canopy by a weather station and then used for canopy evaporation estimation. This estimation was based on two approaches: the slope of the linear regression of gross rainfall against interception, or the Gash procedure (Ev1), and the Penman–Monteith equation (Ev2). Two rainfall interception models (Liu and Gash) were employed, using both Ev1 and Ev2. Thirty-two fixed rain gauges were used for Tf measurements and 32 trees were selected across species and diameter at breast height ranges for Sf observations. The revised version of the analytical Gash model underestimated I by -17.5% and -11.1% for Ev1 and Ev2, respectively, resulting in less accurate estimates. Based on these relative errors, the performance of the Gash model was classified as “fair.” The Liu model overestimated I by 5.3% and 11.3% for Ev1 and Ev2, resulting in assessments of “good” and “fair” respectively, and thus indicating improved performance compared to the Gash model. Therefore, the Liu model coupled with Ev1 is preferable for simulation of rainfall interception in semi-deciduous forests of the BAFB. However, slight overestimation bias was observed, and the model requires tweaking with respect to the estimation of forest storage capacity for this ecosystem. Interception loss modelling is a strategic tool for assessing the influence of different weather patterns and forest vegetative features on water balance in semi-deciduous forest ecosystems.