In this answer I assume that you want to identify the parameters of a model for which you only have the input and output data and perhaps the order of the system; if this is not the case please expand your question.
You mean you want to test an algorithm in simulated data before using it in real data? If this is the case, then the best practice is to simulate a model in Simulink (I'm assuming you are using Matlab) where you have control of the type of model, the input used to stimulate your system and your sampling rate. Then you must take your input and output data, add some realistic noise (sometimes white Gaussian noise is not a realistic representation of the noise found in the system) to your output data and then use your algorithm, the input and noisy output signals to estimate the parameters of the model. In this way you will be able to tell if the algorithm can estimate the true parameters or not.
This practice is recommended because if you go ahead and use your algorithm with real data you will never know if you are getting biased results or not.
For parameter identification you can also refer the help from the ident tool box of matlab which will useful to you to identified the model and test the model derived from the input output data. To generate the test you can also used ode solver of the Matlab or as per diego says.
Of course matlab is the best software to simulate an experimental benchmark, so I think, you may generate a similar data like your plant to improve your identification system for the next step.