The CABRI facility is an experimental nuclear reactor of the French Atomic Energy Commission (CEA) designed to study the behavior of fuel rods at high burnup under reactivity initiated accident conditions, such as a control rod ejection. The distinctive feature of this reactor is its reactivity injection system. The power can rise from 100 kW to 25 GW in a few milliseconds. To know the energy released into a test rod, it is necessary to access the driver core power online. The neutron flux is measured online by compensated boron chambers. These neutron detectors are calibrated during the commissioning phase thanks to standards given by a conventional heat balance. The boron chamber signal depends on the temperature of the pool and the magnitude of the core power according to a nonlinear multivariate model. The uncertainties of the standards and those of the neutron chamber signal cannot be neglected. Moreover, the size of the sample is very small due to the operational constraints. A classic regression method does not take into account all these parameters. In such a situation, we show how the statistical bootstrap method can prove to be a useful and easy tool in tackling this issue. This paper describes first the adjustment of the calibration model that will be used for the prediction during the core power transient and second how we take into account both the uncertainties of the physical variables and the small size of the experimental sample.