This paper investigated the use of spectra and multispectral images for on-site visualized classification of transparent hazards and noxious substances (HNS), such as benzene, xylene, and palm oil, floating on a water surface with the potential use for rapid classification of multiple HNS during a leak accident. Partial least-squares discrimination analysis (PLS-DA) and least-squares support vector machine (LS-SVM) models achieved a classification accuracy of 100% for spectral reflectance (325-900 nm) and multispectral image at nine wavelengths. Wavelength division and selection were applied for spectra and spectral images, respectively, to reduce the difficulty in data collection and to simplify the redundant bands. This was followed by PLS-DA and LS-SVM modeling. The LS-SVM model based on the least wavelengths (365, 410, 450, and 850 nm) of multispectral images was suggested as the most effective method for on-site visualized classification of transparent HNS on a water surface.