Mineral processing relies heavily on the differential behaviour of particles in slurries. In order to monitor the performance of separation units such as flotation cells or hydrocyclones it is helpful to develop on stream characterization tools providing information on particle size distribution and mineralogy. This paper describes an experimental set-up designed for the multispectral imaging of particles in water suspensions. The imaging system enables the acquisition of multispectral images. Images are processed to provide analysis on hundreds of particles in a few seconds. The prototype was used to acquire images of a series of pure mineral samples. Pure minerals powdered samples were suspended in water in order to mimic mineral slurries. The data are processed by means of machine learning methods. It is demonstrated that selected minerals can be discriminated from one another based on their optical properties. It is also shown that the optical properties measured with the system are somehow related to well-documented optical features such as specular and diffuse reflectance. The technology developed in this work sets the basis for on line monitoring of ore slurries with reasonably simple mineralogy.