In this paper a comparative study on the use of extended and unscented Kalman filters for state estimation in nonlinear systems is presented. This is done to reveal the differences, and congruencies, in filters' synthesis in view of real plant implementation on embedded platforms. One of the objectives is to validate the usual performance measures of the filters, meant by state estimation, experimentally. Design features, such as the complexity of implementation and the sensitivity of the state initialization process on uncertain covariance information at regular filters startup are evaluated in a variety of real plant scenarios. A pilot plant of batch distillation process has been used for experimentation. © 2011 IEEE.