To define parameters for optimizing automated discrimination of bacterial biochemical reactions certain theoretical considerations of spectrophotometric analysis were explored. One-hundred and one recent clinical isolates of gram-negative bacilli (21 species) were inoculated into AP1 20 E strips and read manually after 24 hours. With spectrophotometric scanning, the AP1 reactions could be classified into three analytical categories: pH change, production of new products, and darkening of the medium. Whereas single wavelength analysis gave 2.9% disagreement from the visual, multiple wavelength analyses were uniformly more accurate. The best results for pH change reactions were obtained by calculating a ratio of two wavelengths. New color reactions were best interpreted by demonstration of the new peak, whereas darkening reactions required quantitation of the area under the entire curve. With these methods, a 99.3% overall agreement of individual reactions and a 97% agreement of identification were achieved. Multiple-point analysis of spectra coupled with computerized interpretation of the data should help resolve the problem of equivocal reactions in bacterial identification schemes optimized for spectral analysis.