In a sensor network, the technique that limits the number of sensors used for observation is effective to reduce the energy consumption of each sensor. To limit the number of sensors without sacrificing observation accuracy, an appropriate sensor combination must be selected by evaluating the observation effectiveness of various combinations. However, the computational workload for evaluating all the sensor combinations is quite large. We can define a parameter related to the optimal size of a region around an observation target by making a trade-off between accuracy and the computational workload. In region-based sensor selection, a combination of sensors is selected from that is near the observation target. Accuracy is better in a larger region with a lot of sensors, but the computational workload is heavier. In contrast, a smaller region with fewer sensors has poorer accuracy, but a lighter workload. The size of the region controls the trade-off between accuracy and the computational workload. We define a parameter related to the optimal size of a region, and use it to dynamically adjust the region's size. Our simulations confirmed that region-based sensor selection reduces the computational workload and improves accuracy in comparison to existing techniques.