The active K2V star \epsilon Eri hosts the most nearby known extrasolar planet .
With an angular separation of about 1 ” on average , and an age of a few to several hundred Myrs , \epsilon Eri b is one of the prime candidates for becoming the first definitive extrasolar planet imaged directly .
We present a multi-epoch deep differential imaging survey performed with NACO-SDI at the VLT with the aim of finding the planet .
The results are combined with recent astrometry in an attempt to further constrain the detection limits .
No convincing candidate is found among the many coherent structures that constitute the residual speckle noise , which is the dominant noise at small angular scales .
We present our detection limits , compare them with the estimated brightness of \epsilon Eri b , and analyze how the limits can be improved further .
It is found that integration time remains a very important parameter for achieving good results , even in the speckle-dominated regimes .
The results yield new , improved upper 3 \sigma limits on the absolute H-band ( 1.6 \mu m ) brightness of the 1.55 M _ { jup } companion of 19.1 to 19.5 mag , depending on the specific age of the system .