It is well known that the maximum likelihood (ML) algorithm is characterized by their complexity, which is very high. This paper presents a low-complexity semi-blind algorithm to jointly estimating the Carrier Frequency Offset (CFO) and channel based on oversampling in the receiving part for an Orthogonal Frequency Division Multiplexing (OFDM) system. The idea is to substitute the exponential term in the cost function by the first order of Taylor series approximation and to estimate the CFO iteratively until obtaining a low residual value. In addition to demonstrating the effectiveness of our algorithms, we derive the Cramer-Rao bound CRB.