This study presents a simple algorithm named footprint analysis to predict risk and onset of paroxysmal atrial fibrillation (PAF) from single channel surface electrocardiogram. The approach is based on hypothesis that subject who has risk of PAF would display a specific pattern in change of heart rate and could be applied as a predictor to identify the risk and onset of PAF. To quantify patterns of heart rate dynamics, the triplet codes (0, 1, 2) is assigned according to difference of adjacent R-R intervals to represent equal, acceleration, and deceleration of successive RR intervals, respectively. A R-R interval series is then continuously weighted by a 7 heartbeats window, which results in a set of 6-bits number. Each number represents a specific pattern of heart rate dynamics. The strategies for both events of competition are to determine which number has higher possibility to be present in PAF patients, and to identify the number that only present before onset of PAF. A set of number determined by the algorithm from learning dataset was then applied as footprint to the testing dataset. Score for event 1 was 33/50 (entry 20010422.030701) and score for event 2 was 38/50 (CinC Challenge 2001 entry 20010423.045638, entrant 2). The successful of this algorithm is to find hidden pattern embedded in the highly dimensional phase space of R-R intervals. We look forward to understanding the link between microscopic variation in R-R intervals and macroscopic physiologic conditions.