Parameter tuning is an important step in automatic fuzzy model identification from sample data. It aims at the determination of quasi-optimal parameter values for fuzzy inference systems using an adequate search technique. In this paper, we introduce a new hybrid search algorithm that uses a variant of the cross-entropy (CE) method for global search purposes and a hill climbing type approach to improve the intermediate results obtained by CE in each iteration stage. The new algorithm was tested against four data sets for benchmark purposes and ensured promising results.