Monitoring industrial systems in real time raises some cognitive problems especially due to the important number of alerts that must be aggregate by the monitoring system itself and prioritized by operators. Among other senses (vision and sound), color may bring valuable insights to operators for quickly understanding alerts and appropriately react. In our case of supervising interfaces in real time of oil exploitation, the operators have safety constraints concerning environment, machines and human. It is necessary to bring into focus the situation quickly to the operator so that he/she acts. In order to assist the operator during his/her activity and to limit his/her cognitive effort, the supervision interfaces are equipped with alerts which have several important levels like "danger", "critic", "weak anomaly" or "no problem". We perform an experiment focuses on the comparison of three alert colour tested on a supervision interface by users used to this interface types. Our results prove that one alarm color code is 47% more efficient in terms of time reaction of testers. This best colour is a "red gradation" i.e. from bright red for the "danger" alert to white for the "no problem" alert. This alert type is not only efficient for the supervision, but also for a strategic or piloting dashboard, even personal dashboard as personal expenses management.