A number of behavioral and cognitive functions of brain differ between male and female. Occurrences of psychiatric disorders, e.g., attention deficit hyperactivity disorder, autism, depression and schizophrenia also vary from male to female. Understanding the unique cognitive expressions in gender-specific brain functions may lead to insights into the risks and associated responses for a certain external simulation or medications. Previously resting-state functional magnetic resonance imaging (r-fMRI) has been used extensively to understand gender differences using functional network connectivity analysis. However, how the brain functional network changes during a cognitive task for different genders is relatively unknown. This paper makes use of a large data set to test whether task-fMRI functional connectivity can be utilized to predict male vs. female. In addition, it also identifies functional connectivity features that are most predictive of gender. The cognitive task-fMRI data consisting 475 healthy controls is taken from the Human Connectome Project (HCP) database. Pearson correlation coefficients are extracted using mean time-series from anatomical brain regions. Partial least squares (PLS) regression with feature selection on the correlation coefficients achieves a classification accuracy of 0.88 for classifying male vs. female using emotion task data. In addition it is found that inter hemispheric connectivity is most important for predicting gender from task-fMRI.