We consider techniques to remove contaminants when calculating the 3D galaxy power spectrum. The commonly used naïve template subtraction technique estimates the power spectrum quickly but in a biased way. The mode deprojection scheme is designed to give an unbiased power spectrum, but it requires huge costly matrix operations, which make it unpopular to use. We show that, prior to normalisation, template subtraction and mode deprojection yield identical results. We therefore argue that one can remove contaminants from the 3D galaxy power spectrum using template subtraction, and then removes the bias either using the normalisation of the quadratic maximum likelihood (QML) estimator as one would do for mode deprojection, or to use a new and computationally cheaper method that is an extension to the FKP estimator and also debiases the power spectrum estimate, but slightly increases the error bars by not weighting each mode optimally. However, the increase of the error is small in the test cases that we analyse, and is less severe than many other sub-optimal assumptions commonly employed.