Long time series of underwater images have become a tool widely used within the benthic ecology research community. The development of new acquisition systems with bigger storing capacities lead researchers and scientists to deploy them for longer periods resulting in large amounts of data. This paper focuses on the first steps of analyzing large numbers of underwater images, which involves assessing the amount of valid data while assuming no technical problems. The question here addressed is how many of the in situ images can reliably be really used for benthic ecology purposes. To answer this question, we propose a method to eliminate nonvalid images and use it with four different sets of time-lapsed images acquired for long periods ranging from 73 to 371 ds in a row. The results show that elimination of between 8% and 22% of the images is possible depending on the data set. The main advantage of the method is easing and accelerating automation of subsequent analysis.