In this paper, we analyze the performance of IMDD-NRZ ASE-noise-limited systems using MLSE receivers in the presence of CD and DGD. We study the performance of different approximated branch metrics and compare them to the optimum metric evaluated using the Karhunen-Loeve (KL) method. We specifically concentrate on the SQRT metric, because of its low complexity and potential ease of channel estimation. We test it at 10.7 Gb/s, up to about 12000 ps/nm of CD (700 km of SSMF) and about 150 ps of DGD, showing that the SQRT metric performance is virtually coincident with that of the optimum metric. This result holds true over a broad range of TX and RX configurations, using various different filtering optimization strategies. The effects of a finite resolution of the A/D converter on all considered metrics are also analyzed, showing the impact of quantization over the SQRT metric to be the same as for the optimum metric. Finally, the statistical properties of the SQRT metric are discussed. In particular, the probability density functions of signal and noise related to the SQRT metric are compared with the exact ones.