For an unforced nonlinear single output system, we propose a new approach to observer error linearization called reduced-order dynamic observer error linearization (RDOEL), which is a modified version of dynamic observer error linearization (DOEL). While RDOEL also has an auxiliary dynamics and a generalized output injection like DOEL, it offers a lower dimensional observer than DOEL, and allows a complete constructive algorithm in contrast to DOEL when the auxiliary dynamics is a chain of integrators. For RDOEL whose auxiliary dynamics is a chain of integrators (RDOELI), we provide a complete constructive algorithm which presents not only a coordinate transformation between the original system and a nonlinear observer canonical form, but also the minimum number of integrators needed to perform RDOELI. Moreover, we show that if the original system is of dimension n, then the minimum number of integrators is less than or equal to n –- 2. In order to describe our algorithm well, we present an example that the auxiliary dynamics is a chain of two integrators.