We propose a new method of realizing visual object recognition by combining hierarchical feature extraction and temporal binding. In a first time, our system uses a convolutive hierarchical neural network similar to HMAX (Riesenhuber, 1999) to extract features from the input image. Feature extractors are learned using unsupervised learning methods (Kavukcuoglu, 2009). However, instead of using conventional classification methods like SVM, we use a spiking neural network to realize temporal binding association as proposed by (Milner, 1974) and (Malsburg, 1981). Features of the input image are matched to a reference image using a modified version of the model proposed by (Pichevar, 2006). Similar object parts will synchronize to realize recognition. The advantage of using such architecture, as opposed to classical classification methods, is that the local organization of the features is conserved, yielding a more robust recognition. The system is also able to recognize objects from very few training samples. Another advantage of our model is that the matching process with a reference image can help interpolate missing or occluded part of the objects.