Software cost estimation plays a critical role in software project management, as inaccurate predictions can lead to budget overruns, resource shortages, or project failure. Over the years, several approaches have been proposed to improve estimation accuracy, with machine learning and data mining techniques attracting significant attention. Among these, the Multi-Layer Perceptron (MLP) neural network is widely applied due to its ability to capture complex nonlinear relationships. However, the predictive performance of MLP is highly sensitive to hyperparameter settings, making proper optimization essential. This study employs the Grid Search technique to optimize MLP hyperparameters for software cost estimation. Experiments were conducted on six benchmark datasets frequently used in this domain: Desharnais, Maxwell, Kemerer, Albrecht, COCOMO81, and COCONASA. Performance was evaluated using three well-established metrics: MMRE (Mean Magnitude of Relative Error), PRED(0.25), and EF (Error Function). The results demonstrate that Grid Search-based optimization significantly improves MLP performance, reducing MMRE while increasing PRED(0.25) and EF across all datasets. These findings highlight the importance of systematic hyperparameter tuning in enhancing model accuracy and reliability, and they confirm the potential of optimized MLP models for practical software cost estimation tasks