High-intensity transient signals (HITS) detected by transcranial Doppler (TCD) ultrasound may correspond to artifacts or to microembolic signals, the latter being either solid or gaseous emboli. The goal of this study was to assess what can be achieved with an automatic signal processing system for artifact/microembolic signals and solid/gas differentiation in different clinical situations. The authors studied 3,428 HITS in vivo in a multicenter study, i.e., 1,608 artifacts in healthy subjects, 649 solid emboli in stroke patients with a carotid stenosis, and 1,171 gaseous emboli in stroke patients with patent foramen ovale. They worked with the dual-gate TCD combined to three types of statistical classifiers: binary decision trees (BDT), artificial neural networks (ANN), and support vector machines (SVM). The sensitivity and specificity to separate artifacts from microembolic signals by BDT reached was 94% and 97%, respectively. For the discrimination between solid and gaseous emboli, the classifier achieved a sensitivity and specificity of 81% and 81% for BDT, 84% and 84% for ANN, and 86% and 86% for SVM, respectively. The current results for artifact elimination and solid/gas differentiation are already useful to extract data for future prospective clinical studies.