In our previous work, we proposed a calibration method for deep learning (DL) to mitigate the effects of acquisition-related data mismatches in the context of tissue characterization. We showed that the “setting” transfer function can transfer deep learning models between imaging settings. We now extend the calibration method to transfer deep learning models between ultrasound machines. This can lead to reduced cost of model development and also improved understanding of the issues related to the security of deep learning based algorithms in biomedical ultrasound imaging. We gathered four datasets from three different scanners: (i) a SonixOne Machine with an L9-4 transducer, (ii) a Verasonics Vantage 128 Ultrasound Machine with an L9-4 transducer using line by line acquisition, (iii) a Verasonics Vantage 128 Ultrasound Machine with an L9-4 transducer using plane wave compounding, and (iv) a Siemens Antares Ultrasound Machine with an VF10-5 transducer. We used the first dataset as training data and the other datasets as testing data. The DL algorithm learned how to classify two tissue mimicking phantoms. The classification accuracy jumped to 90% from 50% for the second dataset, 70% from 50% for the third dataset, and 61% from 56% for the fourth dataset after incorporating the calibration method. Therefore, the results confirm that a transfer function approach can be used to transfer learning models between scanners for the purpose of classifying materials based on ultrasonic backscatter.