Negative campaigning has become a prevalent campaign strategy not just in the U.S., but also in other established democracies. While negative campaigning has been a prominent focus of the academic literature, the state of knowledge is still mostly based on observational data, often artifacts of campaigning such as content analysis of press releases, campaign ads, or social media posts. Based on a pre-registered conjoint experiment embedded in surveys of more than 800 candidates running in German state elections, the paper aims to explain under what conditions candidates attack their opponents. Rational-choice considerations matter, as candidates are more likely to attack when they see a net gain in the strategy. However, the characteristics and behavior of the opponent also play an important role. Negative campaigning is more likely if the opponent is male, ideologically distant, and has attacked before. In contrast, the closeness of the race and the likelihood of retaliation have no influence on attack behavior. Furthermore, the decision to attack their opponent is largely independent of candidates’ own incumbency status, gender, or personality. By integrating relevant factors that were identified in the literature in one research design, the paper sheds light on the drivers of campaign negativity and points towards the role of further situational factors that are shaping candidates’ behavior on the campaign trail. Beyond negative campaigning, this study demonstrates the value of embedding experimental designs in samples of political elites.