This paper is concerned with the fault detection (FD) in networked systems (NSs). A new model is established by using two Bernoulli stochastic variables to describe the phenomena of one-step random delays and packet losses in data transmission. A new linear filter is designed in the linear minimum variance sense using the innovation analysis approach. Based on the proposed filter, an improved weighted sum of squared residuals (IWSSR) fault detection method is presented. When the packet loss rate is low, it has an effective performance, but it may lose the effectiveness if the packet loss rate is high because the innovation has lost the normal distributed property. To solve this problem, a norm-bounded (NB) fault detection method is also proposed, which avoids the restriction that requires innovation satisfying the normal distribution. In the simulation, the performances of the proposed IWSSR and the NB fault detection methods are verified.