Dental disease is largely preventable and closely linked to poor toothbrushing behaviors. Motion-sensors, such as accelerometers, gyroscopes, and magnetometers, allow for mon- itoring of toothbrushing behaviors. Researchers have attempted to infer tooth surface coverage using sensors attached to the toothbrush handle or embedded in smartwatches. However, the inferences may be deficient because the datasets were collected under structured toothbrushing assumptions performed in con- trolled laboratory settings and not the free-form and irregular brushing patterns observed in real-world settings. To address the aforementioned problem, we collected a dataset of 187 brush- ing sessions, including free-form brushing. We present, to our knowledge, the first motion-sensor dataset obtained during free- form brushing. Using our experiences, we discuss the challenges of studying toothbrushing behaviors in naturalistic settings. We also propose a three-stage method (i.e. pre-processing, brush transition time detection, and time-series classification) to detect the teeth surfaces brushed during a session. Our findings are two-fold: (a) the classification of teeth surfaces during free- form toothbrushing is more challenging than during brushing in controlled settings; (b) high classification accuracy can be achieved using random train-test split of the data (i.e. k-fold cross-validation); however, generalization beyond the participants in the training set poses difficulties. Beyond publishing the first dataset of free-form toothbrushing, we validate our findings by applying our proposed method to our provided dataset, as well as the datasets of toothbrushing in controlled settings.