Rapid detection of foodborne pathogens is one of the most effective ways to overcome food safety problems. To realize a rapid, efficient and label-free detection and classification of foodborne pathogens, this study aims to improve the performance of existing optical fiber confocal backscattering spectrum system. Through this process, the light field diameter is reduced to fit small biological samples, and single spectrum level detection can be achieved. Furthermore, the backscattering micro-spectrum of three categories of common foodborne pathogens (Salmonella enteritidis, Escherichia coli, and Salmonella typhimurium) with similar morphology is measured without labels. A multivariate analysis model is established by combining principal component analysis (PCA) and fuzzy cluster analysis (FCA) at the characteristic wavelength range of 500--800 nm. Results show that the top five principal components contain 80.41% characteristic spectral information. The scores of the top five principal components are taken as the variables for the FCA. The accuracy of 100%, according to the degree matrix of membership, is achieved for the clustering results of three kinds of bacteria. Also, results show that optical fiber confocal backscattering spectroscopy, combined with PCA and FCA, can be used to analyze and classify a single spectrum rapidly, efficiently, and without labels.