Similarity join, an operation that finds all pairs of similar objects in a large collection of objects, is widely used to solve various problems in many application domains including Data Cleansing and Integration, Information Retrieval, Collaborative filtering, Clustering, Pattern Recognition, Bio-informatics, and so on. Existing similarity join algorithms use an inverted index with filtering techniques to avoid unnecessary similarity computation. However, they are inefficient in filtering out dissimilar pairs, especially when element weights must be considered. We contrived an efficient algorithm for similarity joins over weight vectors. It is easily extendable to other similarity predicates that are based on aggregate weighted similarity functions. Our algorithm is mostly based on All-pairs and improved its filtering performance by computing tight similarity upper bounds with little overhead.