This paper proposes a novel approach to coal flow estimation in pulverized coal mills, which utilizes measurements of oxygen content in the flue gas. Pulverized coal mills are typically not equipped with sensors that detect the amount of coal injected into the furnace. This makes control of the coal flow difficult, causing stability problems and limits the plant's load following capabilities. To alleviate this problem without having to rely on expensive flow measurement equipment, a novel observer-based approach is investigated. A Kalman filter based on measurements of combustion air flow led into the furnace and oxygen concentration in the flue gas is designed to estimate the actual coal flow injected into the furnace. With this estimate, it becomes possible to close an inner loop around the coal mill itself, thus giving a better disturbance rejection capability. The approach is validated against a fairly detailed, nonlinear differential equation model of the furnace and the steam circuit using data measured at a Danish power plant.