Many methods for the detection of oil pollution on the sea surface from remotely sensed images have been developed in recent years. However, because of the diverse physical properties of oil on the sea surface in the visible wavelengths, such images are easily affected by the surrounding environment. This is a common difficulty encountered when optical satellite images are used as data sources for observing oil slicks on the sea surface. However, provided the spectral interference generated by the surrounding environment can be regarded as noise and properly modeled, the spectral anomalies caused by an oil slick on normal sea water may be observed after the suppression of this noise. In this study, sea surface oil slicks are extracted by detecting spectral anomalies in multispectral optical satellite images. First, assuming that the sea water and oil slick comprise the dominant background and target anomaly, respectively, an RX algorithm is used to enhance the oil slick anomaly. The oil slick can be distinguished from the sea water background after modeling and suppression of inherent noise. Next, a Gaussian mixture model is used to characterize the statistical distributions of the background and anomaly, respectively. The expectation maximization (EM) algorithm is used to obtain the parameters needed for the Gaussian mixture model. Finally, according to the Bayesian decision rule of minimum error, an optimized threshold can be obtained to extract the oil slick areas from the source image. Furthermore, with the obtained Gaussian distributions and optimized threshold, a theoretical false alarm level can be established to evaluate the quality of the extracted oil slicks. Experimental results show that the proposed method can not only successfully detect oil slicks from multispectral optical satellite images, but also provide a quantitative accuracy evaluation of the detected image.