In association with the potential risks of work accidents in the implementation of construction work, the knowledge of K3 on a construction project is now a fundamental requirement. Aspects of K3 itself will not be able to be run as it should without any intervention from management in the form of a planned effort to manage it (safety management), which is often called the Occupational Safety and Health Management System (SMK3). This research aims to know effects of the knowledge of K3 on the behavior of construction workers as viewed from several aspects related to K3 such as the influence of knowledge variables K3 together to the behavior of construction workers, the effect of knowledge variables K3 partially on the behavior of construction workers and the effect of worker's behavior on work accident.This research were conducted in four construction companies which executed works given by the local government of Luwuk Banggai Regency in 2016. The research approach used in this research is survey method and regression and correlation analysis method having samples of 129 workers. Based on statistical test of hypothesis test, the test result of each K3 knowledge variable shows that all the variables tested (5 K3 knowledge variables ie devinition and initiation K3/X1, K3/X2 management system, personal protector K3/X3, facilities and infrastructure K3/X4, K3/X5 risk) has a strong correlation to worker behavioral variables (Y1). This is evidenced by the correlation numbers R> 0.8 and the determination (r2)>0.7. However, based on the results of simultaneous test (F test) to know the correlation of the five variables of K3 knowledge together to the variable of worker behavior (Y1), the result of statistical test on the simultaneous test shows that in t test (partial) only 2 K3 Good correlation to worker behavior (Y1) that is variable of personal protective device (X3) and K3 risk (X5). According to these two test, it revealed that linear multiple regression result in an equation: Y=0,318X3+0,557X5+1,613 with determinant coefficient (R2) = 71.4 % which means that the model resulting having very good performance.