The present invention discloses a method and a system for building a human fall detection model. The method can include: collecting the speed information of one and more human fall samples in the human fall process; the speed information includes resultant acceleration, speed and time; analyzing the characteristics of the speed information, calculating and extracting low acceleration threshold value a0, resultant acceleration series within fixed time I0, high acceleration threshold value a1, curve area threshold value ΔS and low acceleration time threshold value ΔT, and building the fall detection mechanism. The system of the present invention fully takes into account the characteristics of human body's movement behaviors, so as to increase the detection rate and reduce the misjudgment rate.