This randomized controlled trial or "true experiment" examines the effects of a school garden intervention on the science knowledge of elementary school children. Schools were randomly assigned to a group that received the garden intervention (n?=?25) or to a waitlist control group that received the garden intervention at the end of the study (n?=?24). The garden intervention consisted of both raised-bed garden kits and a series of 19 lessons. Schools, located in the US states of Arkansas, Iowa, Washington, and New York, were all low-income as defined by having 50% or more children qualifying for the federal school lunch program. Participants were students in second, fourth, and fifth grade (ages 6-12) at baseline (n?=?3,061). Science knowledge was measured using a 7-item questionnaire focused on nutritional science and plant science. The survey was administered at baseline (Fall 2011) and at three time points during the intervention (Spring 2012, Fall 2012, and Spring 2013). Garden intervention fidelity (GIF) captured the robustness or fidelity of the intervention delivered in each classroom based on both lessons delivered and garden activities. Analyses were conducted using general linear mixed models. Survey data indicated that among children in the garden intervention, science knowledge increased from baseline to follow-up more than among control group children. However, science knowledge scores were uniformly poor and gains were very modest. GIF, which takes into account the robustness of the intervention, revealed a dose--response relation with science knowledge: more robust or substantial intervention implementations corresponded to stronger treatment effects.