A feature vector is a numerical representation of an object’s characteristics, used as input to machine learning models or for comparing objects. Each position in the vector corresponds to some measurable property — a feature — of the thing being represented.
The term predates modern embeddings and is more general. Classic feature vectors were often hand-engineered: to represent a house you might build a vector of its size, number of rooms, age, and price. Embeddings are a specific, modern kind of feature vector where the features are learned automatically by a neural network rather than chosen by hand, and are not individually interpretable.
In the context of vector databases, the vectors being stored and searched are feature vectors in this broad sense — most commonly learned embeddings. Understanding the term helps connect vector search to the wider history of machine learning, where representing data as vectors of features has long been the foundation for classification, clustering, and similarity comparison.