In Northwest Vietnam, intensification of agricultural production, especially the widespread monoculture of maize, has negatively impacted the environment, triggering land degradation and landslides, harming households’ livelihoods. Agroforestry has emerged as a potential solution to achieve sustainability goals and reverse the current degradation trends. The thesis aims to quantify the sustainability improvement in agricultural production of households in Northwest Vietnam with agroforestry. The two objectives are hence i) to understand the diversity of households driving the impacts of agroforestry and ii) to quantify the impacts of agroforestry in different sustainability dimensions in different types of households. To accomplish the two objectives, the socioeconomic and biophysical heterogeneity in the region was described by combining the Principal Component Analysis with agglomerative clustering to reduce the high variability of the dataset and classify households into clusters. After that, the significant impacts of agroforestry adoption were investigated through sustainability assessment by applying the Propensity Score Matching before calculating the Average Treatment Effect on the Treated. Five clusters of smallholder farmers were found, namely: Off-farm-and-livestock producers, pure off-farm-income-based producers, pure crop-income-based households, fruit-tree-based producers, and coffee-and-tea-based producers. Soil types, land use certification, and distance from home to output markets significantly improve the probability of adopting agroforestry. Being a H’mong, one of the most populated ethnic minority groups in Vietnam, reduces participation of farmers in agroforestry. Observing from the entire population, the agroforestry adoption shows a significant gain in the agricultural income of adopters by 87.67% comparing to non-adopters which is equivalent to 7.96 million VND or 300€ per hectare per year. However, the diversity within types of agroforestry that included market-oriented products could have unexpected negative effect on the food security of households. Indeed, agroforestry increased one more month of insufficient food and a higher level of erosion. Treatment effects at household type level yielded higher off-farm income and lower labor expenditure. The trend aligned with household types that their main income was from non-agricultural activities. But for such clusters indicating high share of farmland for agroforestry as the coffee-and-tea-based producers, their on-farm income did not show any significant change. Thus, this trend implied some hidden costs of adopting agroforestry including the hypothesis that the adopted agroforestry systems were still at the early stage and immature.