Eritrean farmers can cultivate rice by harvesting runoff from >82% available non-agricultural land in agricultural watersheds for crop use and reducing percolation through optimization of tillage. Experiments were conducted with NERICA rice, N11, to optimize irrigation requirements and puddling and compaction to reduce percolation. Experimental field was adjacent to Anseba River at downstream end of the watershed and a pond on the upstream to intercept runoff. Irrigation treatments were runoff farming with maximum runoff application depth of 10 mm (I1), and 50 mm irrigation two (I2), five (I3), and seven (I4) days after disappearance of ponded water in main plots and puddling by one (T1), two (T2) and three (T3) passes of puddler and compaction by three (T4), four (T5) and five (T6) passes of 600 kg roller in sub plots in 3 replications. Soil profile was loam in the surface 0.45 m and coarse sandy loam below forming porous belt. Soil submergence was difficult to maintain, but water table was developed in soil profile due to inflow of seepage from the river and pond. Depth to the water table was within 1.5 ± 0.1 m for >2 months and receded down to 1.7 m by crop maturity. Soil wetness was near field capacity around 0.7 m depth and increased below due to natural sub-irrigation from the water table. Rice roots penetrated 0.8 m in the puddled plots and 0.7 m in the compacted plots. Residual soil moisture of 135-146 mm•m −1 after rice harvesting provides opportunity for planting rapeseed mustered following rice. Puddling was superior to compaction in loam soil. Puddling twice and irrigation 50 mm 7 days after ponded water vanished from surface was sufficient for optimum rice grain yield of 4346 kg•ha −1 and straw yield of 4458 kg•ha −1. Optimum puddling and irrigation schedules reduced crop duration by 6 days without significantly affecting yield. Production function showed that rice grain yield of 4789 kg ha −1 could be obtained by 1009 mm applied water through rainfall and irrigation.