Extra heavy oil (< 10° API) reservoirs in South America contain upwards of 500 Billion barrels of crude oil with few fields on production and several new projects being planned. This paper presents results of a large extra heavy oil (XHO) field, under production since 2001 and developed using long (about 6000 ft.) horizontal wells. Data from more than 600 horizontal wells provided an extensive set for detailed analysis. This study includes production and logging while drilling (LWD) data analysis, detailed mechanistic modeling and field-scale dynamic simulation to improve understanding of production mechanisms and quantify effects of important reservoir parameters on primary production performance of XHO reservoirs. The LWD information was used to determine effective well lengths. The average effective well length was found to be 80% of the main trunk section, and is an important factor that impacts horizontal well performance. Production data from this field were analyzed to obtain initial production rate (IP). The IP show a strong and clear trend with depth, with lower IP for shallower reservoirs, caused by higher oil viscosity (due to lower temperature) and lower pressure in shallower reservoirs. Note that for extra heavy oils, viscosity is a strong function of temperature, unlike typical light oils. A large variation in individual well performance at wells in the same pad (or depth range) is also observed due to variation in reservoir quality and thickness. Detailed dynamic simulation was used to quantify impacts of key uncertainties. We also found that gas production rate may be underestimated and oil recovery may be overestimated in typical field scale XHO models because they may not properly capture pressure and saturation changes in the near-well region. As a result, gas saturations in the near-well region may remain below critical for a longer duration in models using coarse grid blocks impacting forecasts. We recommend using models with finer grids normal to horizontal well trajectory. Learnings from data analysis and mechanistic modeling were validated using a heterogeneous dynamic simulation model. Predicted fluid production rates and reservoir pressure compared well with measured data. This study provides clearer insights into XHO performance. The improved understanding will result in a more reliable production forecast and an optimal development plan, critical for improved assessment or design of new projects.