The deep percolation and actual evapotranspiration from flood irrigation in tropical floodplains were predicted using a numerical model, Hydrus-1D, and a bucket type water balance model. Field experiments were conducted on onion and maize crops grown from December 2015 to May 2016 in small irrigation schemes found in the Lake Tana floodplains of Ethiopia. Experimental fields were selected along a topographic transect to account for soil and groundwater variability. Irrigation volumes were measured using V-notches and irrigation depths (400–550 mm) were calculated, and daily groundwater levels were monitored manually from piezometers installed in the fields. The soil profiles were described at each field and physical properties (texture, FC, PWP, BD, and OM) were measured at each horizon which were used to derive model input parameters. Soil hydraulic properties (residual and saturated moisture content, saturated hydraulic conductivity, parameters related to: pore size distribution n, air entry α and pore connectivity l) were derived using KNN pedotransfer functions for tropical soils and fitted using Retention Curve Program for Unsaturated Soils, RETC. The seasonal actual evapotranspiration estimated by Hydrus and water balance models ranged from 320 to 360 mm for onion and from 400 to 470 mm for maize. The seasonal deep percolation estimated from both models was 12–41% of applied irrigation and with this flood irrigation management; the deep percolation is very high. Implementing precise irrigation and water saving practices that minimize deep percolation and unproductive excessive consumptive use are required to achieve the growing food demand with the available water. When less detailed information is available, the water balance model can be an alternative to predict deep percolation and actual evapotranspiration.