This paper discusses the control design of a magnetically guided microrobotic system in blood vessels to perform minimally invasive medical procedures. Such microrobots consist of a polymer-bonded aggregate of nanosized ferromagnetic particles and a possible payload that can be propelled by the gradient coils of a magnetic device. A fine modeling is developed and used to define an optimal trajectory which minimizes the control efforts. We then synthesize an adaptive backstepping law that ensures a Lyapunov stable and fine tracking, despite modeling errors, and estimates some key uncertain parameters. As the controller synthesis uses the microrobot unmeasured velocity, the design of a high-gain observer is also addressed. Simulations and experiment illustrate the robustness to both noise measurement and some uncertain physiological parameters for a 250-μm radius microrobot that navigates in a fluidic environment.