Attenuation correction is one of the major challenges in the development of PET-MR scanners. Predicting attenuation values from MR images is difficult because MR signals are not related to electron densities and in most MRI sequences, air, bone and lung do not produce any signal, while their attenuation coefficients are completely different. Here we investigate a method to obtain the necessary transmission data simultaneously with the emission data in a existing Time-Of-Flight PET system. An annulus shaped transmission source filled with F-18-FDG is placed inside the FOV of a TOF-PET scanner. A blank PET scan is acquired followed by an acquisition of a phantom containing F-18-FDG. During the second acquisition, photons originating from the transmission source as well as from the phantom are detected. TOF information is used to separate the transmission data from the annulus source and the emission data from the phantom. An iterative gradient descent method is then applied to the transmission data to reconstruct an attenuation map yielding attenuation coefficients at 511 keV. The method was validated with phantom studies on the Gemini TF PET scanner using an anthropomorphic torso phantom and a cylindrical phantom containing bone, lung, water and two hot spheres. The torso phantom contained no activity. Data were corrected for randoms during reconstruction using a delayed window method. A method is presented to derive a global count rate correction factor. No scatter correction was implemented. The reconstructed attenuation maps of both phantom studies show good contrast between different tissues. An underestimation of attenuation coefficients is noticed caused by scattered events. The emission data acquired during the cylindrical phantom study were reconstructed and corrected using the transmission-based attenuation map. A Contrast Recovery Coefficient of 62% and 73% was obtained for the two hot spheres respectively. This work shows how an attenuation map can be reconstructed from a simultaneous transmission/emission scan. In order to obtain accurate attenuation coefficients, corrections for random coincidences, count rate and scatter are needed. The reconstructions presented in this work were only corrected for randoms and the count rate performance mismatch between the blank and transmission scan.