Current generation global climate models suffer from significant biases in the tropics. Specifically, in the tropical Atlantic most climate models fail to simulate a cold tongue in the eastern equatorial ocean during boreal summer in June-July-August (JJA) and exhibit a severe permanent warm SST bias (up to 8~9 degrees C) along the west coast of southern Africa from 15 degrees S to 25 degrees S. Associated with these SST biases are large systematic errors in the simulated regional atmospheric circulation and precipitations over the tropical Atlantic sector. In particular, many models suffer from a southward shift in the Atlantic ITCZ the so-called double ITCZ syndrome and from a dry bias over the Amazon. The Atlantic bias problem has confronted the climate modeling community for decades and little progress has been made to resolve it. This issue has severely undermined the credibility of climate models in simulating and projecting future climate change in the region. In this study, we combine global model analyses with regional climate simulations to shed light on the causes of tropical Atlantic biases. The global model analyses are based on the CMIP5 ensemble, while the regional climate simulations utilize the regional atmospheric model, WRF, and the regional ocean model, ROMS, as well as a coupled WRF-ROMS-CLM model recently developed by our group within the framework of the CESM coupler. Sensitivity experiments with these coupled and uncoupled regional models reveal that small-scale regional features, such as the Angola-Benguela Front, can contribute significantly to the bias problem. Properly resolving these regional features can post a challenge for global climate models.