Falls are the most common reason of injury and one of the leading causes of disability among older people. In order to predict fall risks, it is necessary to evaluate the balance ability. Based on simple tasks, a test and analytical model for dynamic balance and reactive balance is constructed in this paper and multi modal analytical elements are obtained to evaluate the balance ability of the elderly effectively and conveniently. Using motion capture system, 12 subjects are tested for dynamic balance ability and reactive balance ability under normal gait and two kinds of simulated abnormal gait. The two tests are based on normal walking tasks and obstacle-crossing tasks, respectively. By analyzing multi-dimensional motion data quantitatively, four variables are extracted for the nonlinear regression to establish the comprehensive quantitative evaluation model of balance ability. Gait Profile Score (GPS), crossing speed, step length and foot clearance while crossing obstacles are included in the model. The correlation coefficient between the model and the reference value of balance ability score is 0.8898 (p<0.05). Results show that the assessment tasks selected in the experiment are simple and can quantify the balance ability to a single score objectively and effectively, which makes it a practical and feasible model for assessments of balance ability.