We are developing an imaging system to detect pre-/early cancers in the tracheo-bronchial tree. Autofluorescence might be useful but many features remain suboptimal. We have studied the autofluorescence of human healthy, metaplastic and dysplastic/CIS bronchial tissue, covering excitation wavelengths from 350 to 480 nm. These measurements are performed with a spectrofluorometer whose distal end is designed to simulate the spectroscopic response of an imaging system using routine bronchoscopes. Our data provide information about the excitation and emission spectral ranges to be used in a dual range detection imaging system to maximize the tumor vs healthy and the tumor vs. inflammatory/metaplastic contrast in detecting pre-/early malignant lesions. We find that the excitation wavelengths yielding the highest contrasts are between 400 and 480 nm with a peak at 405 nm. We also find that the shape of the spectra of healthy tissue is similar to that of its inflammatory/metaplastic counterpart. Finally we find that, when the spectra are normalized, the region of divergence between the tumor and the nontumor spectra is consistently between 600 and 800 nm and that the transition wavelength between the two spectral regions is around 590 nm for all the spectra regardless of the excitation wavelength, thus suggesting that there might be one absorber or one fluorophore. The use of backscattered red light enhances the autofluorescence contrast.