Nowadays structural health monitoring systems (SHMS) play important roles in assuring the serviceability and safety of some critical infrastructures during their long service lives. The operational performance of SHMS substantially relies on the features of sensor system, including types, number and spatial allocation in structures. Since the number of sensors is always limited compared with degrees-of-freedom of a large-scale structure, the determination of sensor number and locations becomes a critical issue encountered in the design and implementation of an effective SHMS. Meanwhile, the fast development of sensor technology makes various types of sensors available for structural health monitoring purpose, enabling the monitoring of both global behavior and local response. Even though such comprehensive SHMS have been instrumented in many newly-built critical structures, surprisingly little work in the literature focuses on the optimal design of global and local sensors for structural health monitoring. Therefore, this paper attempts to addresses this knowledge gap-the location selection and data fusion of a multi-type sensor system in a structure including displacement transducers, accelerometers and strain gauges, all of which are commonly used in SHMS. The number and locations of the three types of sensors are optimized with the objective of minimizing the estimation error of unobserved structural responses based on incomplete measurement. Unlike traditional approaches for sensor placement in which each type sensors are designed separately, this study designs the whole sensor system simultaneously. By minimizing the overall estimation errors at the locations of interest and reducing estimation errors to a desired target level, the initial set of candidate sensor locations is reduced to a smaller optimal set. Kalman filter algorithm is employed in the sensor placement and data fusion. A numerical examples-a two-dimensional truss structure-was presented to illustrate the effectiveness and accuracy of the proposed approach for the location selection and data fusion of multi-type sensors.