Steam fraction in riser tubes of boilers is a critical process variable which impacts the life of the tubes and could lead to tube rupture, long boiler down time, and expensive repairs. Unfortunately this parameter is difficult to measure by hardware sensors. This article presents a new neural network softsensor for estimation and monitoring steam mass and volume fractions in riser tubes. First, conventional data were collected from a target industrial boiler. The data are then used to develop a detailed nonlinear simulation model for the two phase flow in the riser tubes and risers and downcomers water circulation. The model output is verified against the collected field data. Next, the boiler nonlinear model is used to generate data covering a wide rage of operating conditions for training and testing the neural network. The input of the neural network includes the heating power, the steam flow rate, the water feed rate, the drum level, and the drum pressure. The neural networks predict the mass steam quality and the steam volume fractions. The softsensor achieves a root mean square error on the test data less than 1.5%. The predicted steam quality is then compared with the critical limits to guide the operators for safe and healthy operation of the boilers. The developed softsensor for estimation of the steam quality has simple structure and can be implemented easily at the operator stations or the application servers.