This article presents a mixed methods analysis of 4589 tweets posted between 2012 and 2017 by accounts connected to The Internet Research Agency in St Petersburg, a so-called ‘troll factory’ affiliated with the Russian government. The study departs from a list of 2752 deleted accounts, which Twitter handed over to the U.S. Congress in October 2017 as part of investigations into Russia’s potential meddling in the 2016 U.S. elections. By querying a historical database of tweets as well as the Internet Archive’s Wayback Machine, the study examines a database of nearly 5000 tweets posted by 624 deleted IRA accounts. Tweets span a range geo-political and spatio-temporal contexts, including the Charlie Hebdo terrorist attack in 2015. the Brexit referendum in 2016 and local news affairs in the US from 2014 to 2017. Tweets were manually coded based on 19 variables, developed through an inductive analysis of a sub-sample of tweets. Variables include geo-political context, national identity, endorsements or disapproval of political actors, fear-mongering, populist sentiments, emotional charge, polarization, hostility, conspiracy-theorization, and incitement of offline action. Drawing on propaganda studies, accounts were classified as either white, grey and black propaganda, encompassing identifiable, unidentifiable or disguised sources. The study finds that a majority of accounts were dedicated to disseminating black propaganda (53%, N=333) as opposed to grey (10%, N=62) or white propaganda (37%, N=229). Additionally, accounts with disguised sources (black propaganda) have significantly higher numbers of followers than grey and white propaganda accounts. In contrast, grey propaganda accounts consistently score higher than black and white for fearmongering (x̅=.44, .22, .01, respectively), populist sentiments (x̅=.31, .25, .02, respectively) and hostility (x̅=.26, .15, .01, respectively). The study concludes that propaganda classes can productively account for variances found in the studied data, with black and grey propaganda accounting for a majority of content sowing social discord and antagonism, disseminating polarized information, questioning public safety, and spreading rumors and conspiracy stories. Finally, the article discusses the broader political implications of state propaganda in the age social media, including the difficulties of studying and addressing the phenomenon.