The NOAA Global Monitoring Laboratory (GML) has conducted in situ measurements and discrete air sampling at a network of continental tall towers since the 1990s. Discrete samples are collected automatically using Programmable Flask Packages (PFPs), typically at the same time every day or every few days. The towers reach heights of up to 500 m and are located at elevations ranging from 2.0 to 4200 masl, with daytime sampling within the continental boundary layer. Samples are analyzed in the GML Boulder laboratories for over 50 gases, including long-lived greenhouse gases, ozone depleting substances, and other important trace gases. They are intended to provide regionally representative (O(103) km) measurements of the lower atmosphere composition across North America, including in areas influenced by both natural processes and anthropogenic activities in the continental interior and on both the Atlantic and Pacific coasts. This poster describes a statistical filtering approach for multi-gas PFP flask data, specifically for measurements of CO2, CH4, CO, N2O, SF6, and H2, at 20 tower sites. The primary goal of the filtering method is to accurately identify and flag outlier mole fractions, likely related to local flux processes, while preserving signals representative of regional-scale processes, prior to data release. We describe how this filtering procedure detects anomalous data points and characterizes patterns and typical behavior across the network. The statistical filtering method expands on the NOAA GML curve fitting/smoothing procedure introduced by Thoning et al. (1989) and relies on fitting time series curves for each gas at every site. Outliers are identified using a threshold for residuals relative to the fitted curve. For sites with inconsistent records or significant sampling gaps, the standard curve-fitting approach is modified to use the fitted function directly in place of a smoothed curve. When applying this method to the tall tower PFP dataset, over the full record up to the present, approximately 2–4% of observations are flagged as outliers. We will present examples of filtered and unfiltered datasets, including preliminary results in which outlier identification for a single gas is based on outliers for other gases. Overall, statistical filtering allows for a consistent, documented, and reproducible approach to the identification of outliers in continental data time series.