The adoption of improved varieties and technologies by rural agricultural households remains a major goal in development efforts. This is because researchers and development practitioners recognize it as a potential source of income growth for the poor. However, while studies on adoption of improved technologies abound, little evidence exist on the continued use of improved technologies. This study addresses this gap by focusing on divisible technologies which do not require significantly high investments or capital. A dynamic stochastic model is formulated which analyzes the mechanics of adoption and abandonment of improved technology among poor households. The model is solved using numerical approximation methods for different regimes of financial intermediation. The results show that increasing credit limits helps in sustained adoption and prevents abandonment of high-income crop. Using discrete choice panel data models, we test the implications of our theoretical results with data from rural Ethiopia. The estimation results confirm that for households already engaged in high income crops like maize, access to credit prevents abandonment of the maize crop.