In a pre-registered online experiment, I manipulate the production function used to transform two participants’ work into money. In one treatment participants’ inputs are perfect substitutes, while in the other treatment participants’ inputs are perfect complements. Once the money has been produced, each pair of participants (pairs are formed randomly within each treatment group) have five minutes to decide how it should be divided between them, using an unstructured bargaining protocol. I consider five outcomes of the bargaining process: frequency of disagreement, the agreement point, the distance of the agreement point from a focal point, the number of proposals sent by a bargaining pair, and the bargaining time. The data shows that there is no statistically significant difference in these outcome variables between the two treatment groups. There is a statistically significant increase in the probability of agreeing to an equal division when the production function treats inputs as complements. The interpretation of these findings is affected by a large difference in the attrition rate between the treatment groups. 41.5% of the participants assigned to the treatment with the substitutes production function were excluded from the sample, compared to 25% of the participants who were assigned to the treatment with the complements production function. This large difference in the attrition rate exists because participants assigned to the treatment with the substitutes production function, failed a comprehension test which tested their understanding of the treatment applied to them, with much higher frequency.