The number of continuous operating geodetic stations equipped with a GNSS receiver has increased in the last two decades from a few hundred to several thousands. It is nowadays the main tool for studying tectonic plate motion, glacial rebound of the Earth’s crust due to present and past melting of the ice at Greenland and Antarctica, and the measurement of vertical land motion at tide gauges to separate sea-level rise from sinking of the land. Unfortunately, most GNSS position time series suffer from offsets which have a significant effect on the estimated linear trend, representing the long-term motion. Some of these offsets are due to earthquakes and/or receiver/antenna changes that are often reported in the log files. Nevertheless, presence of unreported offsets with unknown causes require each GNSS position time series to be screened to ensure that all offsets have been detected. Due to the large number of GNSS stations, there is a growing need for developing algorithms towards automatic offset detection. Over the years various offset detection algorithms have been developed. In this research we use the simple approach of testing the effect of adding an offset at each data point by computing how it improves the fit of the model (i.e. linear trend plus seasonal signal and offsets) with the observations. The quality of the fit is quantified using the log- likelihood, the Akaike and Bayesian Information Criteria as well as a modification of the latter. Once a significant offset has been detected, it is added to the model and the process is repeated to search for the next offset until all have been found. The originality of our work lies in the fact that we consider the temporal correlations during the offset detection process, increasing the number of true offsets while keeping the number of false detected offsets low. So far, this aspect has been mostly ignored in literature mainly due to computational speed. Nevertheless, using the fast maximum likelihood method of Bos et al. (2013) we were able to detect offsets in around 9000 time series available through the Nevada Geodetic Laboratory. We will present the results of this analysis.