The use of rainfall data for flood early warning prediction have been highlighted by several researchers, in the last couple of decades. This study investigates the involvement of antecedent daily rainfall, for the determination of rainfall thresholds, to be used for flood early warning purposes at the upper watershed, Java island, Indonesia. An inventory of 70 flood occurrences for the period of 1992–2017 was compiled, and rainfall data were retrieved from 37 stations. First, calculate the critical discharge to determine the flood status in each watershed based on the results of statistical analysis of the frequency of the data series of discharge. Second, a procedure for the calculation of rainfall thresholds for flood occurrence was followed consisting of four steps: i) determining the rainfall associated with each inventory of flood occurrence and nonoccurrence; ii) the antecedent daily rainfall was calculated for 1 to 7 days for the selected dates and watersheds; iii) the optimum number of antecedent rainfall days was evaluated; and (iv) empirical rainfall thresholds for flood occurrence were determined. The results showed flood occurrences are best predicted using a combination of daily rainfall and 7 days of antecedent rainfall for all alert zones (A, B, C and D) including regional model (RRTM) with a negative relationship between antecedent rainfall and daily rainfall. Rainfall threshold models have an overall accuracy of more than 95%. It has provided evidence that the flood event in the study area is preceded by soil conditions that is saturated due to rain a few days before the flood