Herbicides are the most applied pesticides in the world. Despite numerous laboratory studies demonstrating the toxic effect of herbicides on non-target organisms, the effect of herbicides on soil organisms in the field remains complex to understand and is still controversial. In order to understand how changes in agricultural practices aiming to reduce herbicide use could impact soil biodiversity, we studied the effect of the frequency of herbicide application on soil biodiversity in a tropical agroecosystem. Our study was conducted on banana farms in Martinique, an island with a humid tropical climate belonging to the Caribbean biodiversity hotspot. Thirteen banana plots from five different farms were selected, ranging from plots receiving no herbicides to plots receiving 4–5 applications per year. Soil macro-arthropods were sampled using pitfall traps resulting in the collection of over 6,200 individuals. Of the 100 taxa that were differentiated, 75 could be identified to species level which allowed to assign each taxon to a trophic group and when possible to classify them according to whether they were introduced or native. Macro-arthropod mean species richness was 21% lower in plots with the highest frequency of herbicide application. However, no conclusive effect of herbicides on macro-arthropod abundance was demonstrated. Mean species richness for different trophic groups also decreased with herbicide applications with decreases of 22% for predators, 17% for omnivores, 55% for herbivores, and 55% for decomposers in plots with 4–5 herbicide applications per year compared to plots with no herbicide use. Species composition of macro-arthropod communities varied significantly with herbicide applications. More specifically, we found that native species represented a higher proportion of individuals captured in plots where no herbicides were used; suggesting that agroecological practices implemented at the field level to reduce the frequency of herbicide use potenti