We introduce a new technique to quantify highly structured spectra for which the definition of continua or spectral features in the observed flux spectra is difficult. The method employs wavelet transformation which allows the decomposition of the observed spectra into different scales. A procedure is formulated to define the strength of spectral features so that the measured spectral indices are independent of the flux levels and are insensitive to the definition of continuum and also to reddening. This technique is applied to Type Ia supernovae spectra, where correlations are revealed between the luminosity and spectral features. The current technique may allow for luminosity corrections based on spectral features in the use of Type Ia supernovae as cosmological probe.