Dairy cow diets often exceed protein requirements for milk production. At the same time, dietary protein use efficiency for milk production is low leading to excretion of large amounts of nitrogen to the environment. We assessed the effects of feeding two grass/clover silages mainly differing in crude protein (CP) content on milk and component yields, nitrogen and gross feed utilization efficiency with Norwegian Red dairy (NRF) cows. Forty-eight early- to mid-lactation NRF dairy cows were randomly allocated to two dietary treatments (n = 24) after blocking by initial milk yield, stage of lactation, body weight and parity. Cows were fed silages either low in CP (late cut silage, LCPS; 112 g CP per kg dry matter (DM)) or optimal (mixture of 4 different silages, OCPS; 142 g per kg DM) ad libitum for a period of 54 days. These basal diets were augmented with a fixed level of concentrate feed (160 g CP per kg DM). This was estimated using the Nordic Feed Evaluation System (TINE Optfor KU optimization program) assuming OCPS as an available silage. We hypothesized that OCPS would support higher milk yield than LCPS, and the LCPS cows would consume more feed if they were to achieve similar level of milk yield as their OCPS counterparts. Contrary to our hypothesis, milk (23.7 vs. 24.5 kg per day), energy corrected milk (25.2 vs. 25.6 kg per day) yields and milk components were not significantly different between the groups (LCPS vs. OCPS). Furthermore, LCPS cows sufficiently matched DMI from the silage part of the diet to that of the OCPS cows (12.7 vs. 12.4 kg per day). However, OCPS decreased (P < 0.01) nitrogen use efficiency compared to the LCPS (30 vs. 33%). Our results confirm that reduction in dietary CP levels (ca 130 g per kg DM) can be achieved without loss of production, with reduced N excretion to the environment and reduced cost of milk production in moderately yielding cows. This would have a positive contribution to the current efforts being made to base our dairy production on locally available resources with minimal recourse to imported inputs.