Sensors are currently being used for applications in buildings. In sensor grids, a significant amount of sensor data may be lost. This paper tackles the issue of unreliable sensors in buildings. The common sensor faults known in the literature are bias and outliers. Occurrences of data gaps have not been given adequate attention in the research literature. A methodology based on statistical approach allows the automatic thresholding for data gap detection, i.e., abnormalities on the delay for a set of heterogeneous sensors in instrumented buildings. The efficiency of the method is evaluated on measurements obtained from a real building: an office at Grenoble Institute of technology with a large number of sensors.