In many practical applications involving speech recognition it is of great importance to be able to handle noise suppression in the preprocessing stage. In this paper we describe different front-end processing systems, one based on speech production modelling and two based on auditory modelling, and present results on their formant estimation abilities when being excitated by speech signals contaminated by noise. Three noise types - car, cocktailparty and open-plan office noise - are added to speech signals at signal-to-noise ratios varying between 20 and -10 dB. The results show that preprocessing using auditory modelling is much more robust to noise than speech production modelling, and that formants can still be reliably estimated at SNR = -5 dB for speech signal contaminated by car noise.