In this paper a forward–backward basis function approach for instantaneous frequency estimation of the frequency-modulated signal in noisy environment is presented. At first, a forward–backward prediction approach is applied for least squares estimation of time-varying autoregressive parameters. A time-varying parameters are expressed as a summation of constants multiplied by basis functions. Then, the time-varying frequencies are extracted from the time-varying autoregressive parameters by calculating the angles of the estimation error filter polynomial roots. The experimental results are presented, which shows the superiority of the proposed method against the covariance (forward prediction) approach.