Under rainfed conditions in Tarlac, Central Luzon, Philippines, risk was characterized using field and household panel data for 46 rice farmers over the period 1990–93. Measures of risk at the field level were calculated using pooled time-series cross-sectional data. Field-level risk was found to be quite high, with the average coefficient of variation of yield being 45% and 30% for fields with low and moderate elevations respectively. Farmers applied less nitrogen in fields with higher yield variability, and adjusted the quantity of nitrogen in order to reduce losses in poor years and benefit from greater potential in good years. Farmers with higher levels of education had lower variability of rice income than farmers with a lower level of education. Non-crop income helped reduce total income variability, especially for farmers with high variability of crop income. Implications for technology design and policy improvements are derived from these results.