This study investigated an adaptive soft sensor based on nonlinear differential-algebraic observer for chemical process. The primary variables of many chemical processes can not be measured in real time and they can only be corrected online by secondary variables, so the state observer was used to build soft sensor. In this paper, an adaptive state observer for nonlinear differential-algebraic system was proposed, whose output feedback gain matrix could be obtained online by pole assignment of the linearized subsystem. If the linear subsystem from local linearization on each current operating point is observable, the output feedback gain matrix can guarantee the convergence of the state observer. The fluid catalytic cracking unit was used as an example. An adaptive soft sensor based on nonlinear differential-algebraic observer was established to estimate the unmeasurable variables online; and the estimated values of the unmeasurable variables converged to the real values and the soft sensor had good dynamic performance.