ABSTRACT. The paper is aimed at a methodological development of change-point detection, applicable in identifying abrupt changes in temporal or spatial data sequences. In earlier papers we developed a method for detecting a change in the parameters of a discrete distribution, with the simultaneous estimation of the (deterministic but unknown) distribution parameters before and after the change. In this paper we not only extend this method to the case of normal distributions, but also provide a new algorithm for the iterative refining of the estimation of the change-point, based on a "cleaning" of mixed-up parts of the samples. The appropriate size of reduced part of the sample is analytically calculated for the case of normal distributions. This "cleaning" is combined with our original change-point detection method. Our new algorithm is not only validated on artificial data, but also applied to a real environmental data set collected and analysed by other authors in a seafloor observatory. Our results detecting abrupt changes of bacterial mat coverage of a seafloor area are in harmony with the biological fluctuations and changes in the abiotic environment, analysed recently by other authors using a different method. We also provide a comparison with other existing change-point detection methods: a one-dimensional version of the gradient method widely used for edge detection, and a maximum type statistical method well-known in environmental studies. Although normality conditions of our method are rather restrictive, its application potential for environmental data sets is also demonstrated.