Human face identification is a main computational step for many information-processing applications including security checkpoints, surveillance systems, video conferencing, and picture telephony. A new approach is presented for recognizing human faces and discriminating expressions associated with them in color images. It is a statistical technique based on the process of drawing facial silhouettes and characterizing them by autoregressive moving average (ARMA), which, is, in turn, infinite impulse response (IIR) filtering. First, a facial image is transformed from its (R, G, B) space to its principal component representation. A line-drawing profile of the face image is created from its principal component using the zero-crossings of a Laplacian of Gaussian (LoG) filter. The face line-silhouette is then partitioned into 5 × 5 non-overlapping blocks, each of which is filtered by a non-causal IIR filter. The IIR coefficients are approximated by the ARMA parameter vector a. By computing the ensample average of a over the whole image area, we obtain the ARMA feature vector of the facial pattern. Face discrimination is achieved by the non-metric similarity measure S = |cos ∠(a.b)| for two face patterns whose feature vectors (a and b) consist of the aforementioned ARMA coefficients. Experimental results obtained from a small database indicate that the ARMA modeling is capable of discriminating facial color images, and has the ability of distinguishing facial expressions.