A new array‐based method for seismic waveform data is developed to detect small‐scale heterogeneities in the mantle and constrain their possible locations and sizes. Detecting small‐scale heterogeneities using scattered waves has been challenging because scattered phases usually suffer from low signal‐to‐noise ratios, and determining the shapes of such heterogeneities is even more difficult because of isochronal artifacts. In the proposed method, the problem of the low signal‐to‐noise ratio is addressed by adopting dual bootstrap stacking, which can detect weak signals even with a limited number of seismic traces, as well as by ensuring reproducibility among multiple events. The locations of detected scatterers and their sizes and shapes are then estimated using a new kind of hierarchical cluster analysis that could minimize the effect of isochronal artifacts. The new method is applied to a previously analyzed data set of Mariana earthquakes to facilitate a comparison with conventional methods. In the middle to lower mantle beneath the Mariana Trench region, a total of 39 scatterers are identified via S‐to‐P scattering, with their potential volumes ranging from <8 × 104 to ∼1 × 106 km3.