We present time-of-arrival measurements and timing models of 47 millisecond pulsars (MSPs) observed from 2004 to 2017 at the Arecibo Observatory and the Green Bank Telescope by the North American Nanohertz Observatory for Gravitational Waves (NANOGrav). The observing cadence was three to four weeks for most pulsars over most of this time span, with weekly observations of six sources. These data were collected for use in low-frequency gravitational wave searches and for other astrophysical purposes. We detail our observational methods and present a set of time-of-arrival (TOA) measurements, based on "narrowband" analysis, in which many TOAs are calculated within narrow radio-frequency bands for data collected simultaneously across a wide bandwidth. A separate set of "wideband" TOAs will be presented in a companion paper. We detail a number of methodological changes compared to our previous work which yield a cleaner and more uniformly processed data set. Our timing models include several new astrometric and binary pulsar measurements, including previously unpublished values for the parallaxes of PSRs J1832-0836 and J2322+2057, the secular derivatives of the projected semi-major orbital axes of PSRs J0613-0200 and J2229+2643, and the first detection of the Shapiro delay in PSR J2145-0750. We report detectable levels of red noise in the time series for fourteen pulsars. As a check on timing model reliability, we investigate the stability of astrometric parameters across data sets of different lengths. We report flux density measurements for all pulsars observed. Searches for stochastic and continuous gravitational waves using these data will be subjects of forthcoming publications.

We present a new analysis of the profile data from the 47 millisecond pulsars comprising the 12.5-year data set of the North American Nanohertz Observatory for Gravitational Waves (NANOGrav), which is presented in a parallel paper (Alam et al.; NG12.5). Our reprocessing is performed using "wideband" timing methods, which use frequency-dependent template profiles, simultaneous time-of-arrival (TOA) and dispersion measure (DM) measurements from broadband observations, and novel analysis techniques. In particular, the wideband DM measurements are used to constrain the DM portion of the timing model. We compare the ensemble timing results to NG12.5 by examining the timing residuals, timing models, and noise model components. There is a remarkable level of agreement across all metrics considered. Our best-timed pulsars produce encouragingly similar results to those from NG12.5. In certain cases, such as high-DM pulsars with profile broadening, or sources that are weak and scintillating, wideband timing techniques prove to be beneficial, leading to more precise timing model parameters by 10-15%. The high-precision multi-band measurements in several pulsars indicate frequency-dependent DMs. The TOA volume is reduced by a factor of 33, which may ultimately facilitate computational speed-ups for complex pulsar timing array analyses. This first wideband pulsar timing data set is a stepping stone, and its consistent results with NG12.5 assure us that such data sets are appropriate for gravitational wave analyses.