35 Background: The optimal number of cycles of docetaxel for patients (pts) with mCRPC is not known, and in practice, treatment breaks are common. The current study was designed to test the safety and efficacy of utilizing 6 cycles of standard docetaxel with chemo free intervals in patients who achieve and maintain a response to docetaxel. Methods: Pts with mCRPC, no prior chemo, and KPS > 60% were eligible. Pts were treated with “induction” docetaxel 75 mg/m2 q 3 weeks, and prednisone 5 mg po bid. PSAWG1 criteria were used to define response and progression. After 6 cycles, responding pts stopped chemo and were randomized to observation (Obs) or to GM-CSF, 250 mcg/m2 daily for 14 days out of every 28-day cycle. Pts were followed with monthly PSA and imaging every 2 cycles until progression, at which point docetaxel was reinitiated for another 6 cycles, followed by the same “off chemo” regimen. The primary endpoint was the time to progression while on chemo (time to chemo resistance). Results: 114 pts have been enrolled: 3 are undergoing induction, and 111 are therefore evaluable. Of these pts, 82 completed induction, (10 did not due to PD, 9 due to adverse events (AE), 10 due to pt or MD choice). Of 111 evaluable pts, 48 (43%) had a response to chemo and were eligible for randomization. 22 were randomized to Obs and 26 to GM-CSF. Of 48 randomized pts, 25 restarted chemo, all for PSA PD. (23 pts did not re-start chemo because of AE, other therapy being started, or pt choice; 1 pt is still on GM-CSF.) 6/25 (24%) pts experienced a response to the 2nd series of chemo, and 1/6 (17%) to the 3rd. The time to chemo re-initiation (n=25) was 3.1 mos in Obs pts and 4.2 mos in GM-CSF pts. Conclusions: 43% of patients met criteria for undergoing intermittent chemo. The response proportion to the 1st, 2nd, and 3rd series of docetaxel was 43%, 24% and 17%, respectively. GM-CSF may modestly delay the time to chemo re-initiation, but the sample size is small and insufficient to assess the impact of GM-CSF on time to chemo resistance.