The public health surveillance system is one of the most important for the detection of the seasonal influenza epidemic. We introduced different kinds of surveillance data for early detection of a disease outbreak. Hidden Markov model has been recognized as an appropriate method to model disease surveillance data. In this work, we proposed a hidden Markov model (HMM) to characterize epidemic and non-epidemic dynamic in a time series of influenza-like illness incidence rates and presents a method of influenza detection in an epidemic. ILI is defined as an illness marked by the presence of a fever ( 100.5℉ or greater ) and either a cough or sore throat within 72 hours of ILI symptom onset, or physician-diagnosed ILI. ILI incidence rate is based on surveillance data and activity state. HMMs have been used in many areas, including automatic speech recognition, electrocardiographic signal analysis, the modeling of neuron firing and meteorology. A two-state HMM is applied on incidence time series assuming that those observations are generated from a mixture of Gaussian distribution. Bayesian inference method is calculated to obtain the probability of an epidemic state and non-epidemic state every week. The various influenza dataset applied the methodology.