We present a novel approach for 3D shape synthesis based on using fine-grained parts extracted from a set of input models, yielding shapes with a wide variety of fine details. Unlike most previous works, our method does not require a semantic segmentation, nor a part correspondence between the shapes of the input set. After extracting our fine-grained segments, we compute the similarity between these segments using a set of descriptors. Next, we select a template shape among the input models. For each fine-grained part of that template, we use our similarity metric to find a compatible part that we can use for replacing the original part of the template. We show with several experiments that our algorithm can synthesize many distinct shapes by choosing different compatible segments, and by using different templates. Additionally, we maintain the plausibility of the novel objects by preserving the general structure of the template.