In this paper a new deterministic subspace approach is introduced for multi-dimensional nonlinear system identification based on output-only measurements. The strategy is to excite an arbitrary degree of freedom (DOF) and estimate linear and nonlinear parameters of other DOFs within continuous-time state space models. The nonlinearity is characterized in terms of the measured response and its derivatives in locations away from the excitation. Then, basis functions are assigned to predict the nonlinear restoring force and exploited instead of the input data in the identification process. It is demonstrated that projecting the output row space onto these functions, cancels the subspace corresponding to the nonlinear dynamics and the remaining space contains the dynamics of the underlying linear system as well as the contributions of unmeasured input and noise data to the system response. The recent concept and required assumptions are presented through a few mathematical statements and proofs. To ensure the validity of the proposed approach, several discrete and continuous systems with different structural nonlinearities are studied. The estimated parameters are compared with those obtained from traditional subspace identification algorithms and the direct parameter estimation method. The results highlight the robustness of the proposed approach in dealing with noise contaminated measurements.