A task of water supply systems is to provide safe drinking water to every customer, which is a basic human need. Aging of water supply networks and increased precaution of terrorism risks led to re-evaluation of drinking water supply system reliability and vulnerability to accidental and intentional contamination. Contamination of drinking water can cause health, social, psychological and economic issues. During the last decade, early warning systems (EWS) were often used to ensure the safety of drinking water. EWS are driven by conventional sets of drinking water quality sensors, and the collected data are analyzed in real time. For detection of contamination events, numbers of algorithms have been developed. Most of the algorithms are based on statistical analysis or machine learning. The aim of this study was to compare existing methods and to identify the method, which is suitable for contamination detection in drinking water from non-compound specific sensors and requires relatively low computational resource. A detailed review of 11 different algorithms was presented in the current study with the primary focus on detection probability. Cluster analysis in combination with Mahalanobis distances of feature vectors and Canonical correlation analysis (CCA) approach were selected as the most promising methods for application in a new generation of EWS to detect and classify possible contamination events and agents. While canonical correlation analysis method was the most accurate for detection of contamination events, an advantage of Mahalanobis distances was that it not only detects the contamination events but also could identify the type of contaminant. In this study, we conclude that CCA and Mahalanobis distance methods might be applied for detection of contamination events with relatively high and reliable precision.