The invention provides a method and device for monitoring and controlling network risks. The method comprises the following steps: step one, selecting the sample space of network objectives of some network element (NE), carrying out statistic analysis on the sample point data of the sample space, and calculating a dynamic baseline, wherein the network objectives comply with a normal distribution character; step two, according to preset tolerability and the dynamic baseline, determining the upper tolerance limit and the lower tolerance limit of the network objectives as early warning thresholds for triggering an early warning generating mechanism of the predictive network objectives; and step three, judging whether values of the predictive network objectives which are monitored and controlled in real time exceed the early warning thresholds; and if the values exceed the early warning thresholds, triggering the early warning generating mechanism. By realizing a dynamic threshold method, the threshold values can be set more reasonably and accurately, multiple early warning grades can be triggered, accurate monitoring and control for networks is reached, hidden failures in the networks are found, and finally after-event problem analysis of the prior art for the network failures is changed into initiative monitoring and controlling of the network objectives before problems generate, thus effectively ensuring the normal operation of the networks.